xref: /linux/drivers/virtio/virtio_balloon.c (revision 8804d970fab45726b3c7cd7f240b31122aa94219)
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/page_reporting.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 int)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
28 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
29 /* Maximum number of (4k) pages to deflate on OOM notifications. */
30 #define VIRTIO_BALLOON_OOM_NR_PAGES 256
31 #define VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY 80
32 
33 #define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \
34 					     __GFP_NOMEMALLOC)
35 /* The order of free page blocks to report to host */
36 #define VIRTIO_BALLOON_HINT_BLOCK_ORDER MAX_PAGE_ORDER
37 /* The size of a free page block in bytes */
38 #define VIRTIO_BALLOON_HINT_BLOCK_BYTES \
39 	(1 << (VIRTIO_BALLOON_HINT_BLOCK_ORDER + PAGE_SHIFT))
40 #define VIRTIO_BALLOON_HINT_BLOCK_PAGES (1 << VIRTIO_BALLOON_HINT_BLOCK_ORDER)
41 
42 enum virtio_balloon_vq {
43 	VIRTIO_BALLOON_VQ_INFLATE,
44 	VIRTIO_BALLOON_VQ_DEFLATE,
45 	VIRTIO_BALLOON_VQ_STATS,
46 	VIRTIO_BALLOON_VQ_FREE_PAGE,
47 	VIRTIO_BALLOON_VQ_REPORTING,
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 	/* Shrinker to return free pages - VIRTIO_BALLOON_F_FREE_PAGE_HINT */
114 	struct shrinker *shrinker;
115 
116 	/* OOM notifier to deflate on OOM - VIRTIO_BALLOON_F_DEFLATE_ON_OOM */
117 	struct notifier_block oom_nb;
118 
119 	/* Free page reporting device */
120 	struct virtqueue *reporting_vq;
121 	struct page_reporting_dev_info pr_dev_info;
122 
123 	/* State for keeping the wakeup_source active while adjusting the balloon */
124 	spinlock_t wakeup_lock;
125 	bool processing_wakeup_event;
126 	u32 wakeup_signal_mask;
127 };
128 
129 #define VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST (1 << 0)
130 #define VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS (1 << 1)
131 
132 static const struct virtio_device_id id_table[] = {
133 	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
134 	{ 0 },
135 };
136 
page_to_balloon_pfn(struct page * page)137 static u32 page_to_balloon_pfn(struct page *page)
138 {
139 	unsigned long pfn = page_to_pfn(page);
140 
141 	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
142 	/* Convert pfn from Linux page size to balloon page size. */
143 	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
144 }
145 
start_wakeup_event(struct virtio_balloon * vb,u32 mask)146 static void start_wakeup_event(struct virtio_balloon *vb, u32 mask)
147 {
148 	unsigned long flags;
149 
150 	spin_lock_irqsave(&vb->wakeup_lock, flags);
151 	vb->wakeup_signal_mask |= mask;
152 	if (!vb->processing_wakeup_event) {
153 		vb->processing_wakeup_event = true;
154 		pm_stay_awake(&vb->vdev->dev);
155 	}
156 	spin_unlock_irqrestore(&vb->wakeup_lock, flags);
157 }
158 
process_wakeup_event(struct virtio_balloon * vb,u32 mask)159 static void process_wakeup_event(struct virtio_balloon *vb, u32 mask)
160 {
161 	spin_lock_irq(&vb->wakeup_lock);
162 	vb->wakeup_signal_mask &= ~mask;
163 	spin_unlock_irq(&vb->wakeup_lock);
164 }
165 
finish_wakeup_event(struct virtio_balloon * vb)166 static void finish_wakeup_event(struct virtio_balloon *vb)
167 {
168 	spin_lock_irq(&vb->wakeup_lock);
169 	if (!vb->wakeup_signal_mask && vb->processing_wakeup_event) {
170 		vb->processing_wakeup_event = false;
171 		pm_relax(&vb->vdev->dev);
172 	}
173 	spin_unlock_irq(&vb->wakeup_lock);
174 }
175 
balloon_ack(struct virtqueue * vq)176 static void balloon_ack(struct virtqueue *vq)
177 {
178 	struct virtio_balloon *vb = vq->vdev->priv;
179 
180 	wake_up(&vb->acked);
181 }
182 
tell_host(struct virtio_balloon * vb,struct virtqueue * vq)183 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
184 {
185 	struct scatterlist sg;
186 	unsigned int len;
187 
188 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
189 
190 	/* We should always be able to add one buffer to an empty queue. */
191 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
192 	virtqueue_kick(vq);
193 
194 	/* When host has read buffer, this completes via balloon_ack */
195 	wait_event(vb->acked, virtqueue_get_buf(vq, &len));
196 
197 }
198 
virtballoon_free_page_report(struct page_reporting_dev_info * pr_dev_info,struct scatterlist * sg,unsigned int nents)199 static int virtballoon_free_page_report(struct page_reporting_dev_info *pr_dev_info,
200 				   struct scatterlist *sg, unsigned int nents)
201 {
202 	struct virtio_balloon *vb =
203 		container_of(pr_dev_info, struct virtio_balloon, pr_dev_info);
204 	struct virtqueue *vq = vb->reporting_vq;
205 	unsigned int unused, err;
206 
207 	/* We should always be able to add these buffers to an empty queue. */
208 	err = virtqueue_add_inbuf(vq, sg, nents, vb, GFP_NOWAIT | __GFP_NOWARN);
209 
210 	/*
211 	 * In the extremely unlikely case that something has occurred and we
212 	 * are able to trigger an error we will simply display a warning
213 	 * and exit without actually processing the pages.
214 	 */
215 	if (WARN_ON_ONCE(err))
216 		return err;
217 
218 	virtqueue_kick(vq);
219 
220 	/* When host has read buffer, this completes via balloon_ack */
221 	wait_event(vb->acked, virtqueue_get_buf(vq, &unused));
222 
223 	return 0;
224 }
225 
set_page_pfns(struct virtio_balloon * vb,__virtio32 pfns[],struct page * page)226 static void set_page_pfns(struct virtio_balloon *vb,
227 			  __virtio32 pfns[], struct page *page)
228 {
229 	unsigned int i;
230 
231 	BUILD_BUG_ON(VIRTIO_BALLOON_PAGES_PER_PAGE > VIRTIO_BALLOON_ARRAY_PFNS_MAX);
232 
233 	/*
234 	 * Set balloon pfns pointing at this page.
235 	 * Note that the first pfn points at start of the page.
236 	 */
237 	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
238 		pfns[i] = cpu_to_virtio32(vb->vdev,
239 					  page_to_balloon_pfn(page) + i);
240 }
241 
fill_balloon(struct virtio_balloon * vb,size_t num)242 static unsigned int fill_balloon(struct virtio_balloon *vb, size_t num)
243 {
244 	unsigned int num_allocated_pages;
245 	unsigned int num_pfns;
246 	struct page *page;
247 	LIST_HEAD(pages);
248 
249 	/* We can only do one array worth at a time. */
250 	num = min(num, ARRAY_SIZE(vb->pfns));
251 
252 	for (num_pfns = 0; num_pfns < num;
253 	     num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
254 		page = balloon_page_alloc();
255 
256 		if (!page) {
257 			dev_info_ratelimited(&vb->vdev->dev,
258 					     "Out of puff! Can't get %u pages\n",
259 					     VIRTIO_BALLOON_PAGES_PER_PAGE);
260 			/* Sleep for at least 1/5 of a second before retry. */
261 			msleep(200);
262 			break;
263 		}
264 
265 		balloon_page_push(&pages, page);
266 	}
267 
268 	mutex_lock(&vb->balloon_lock);
269 
270 	vb->num_pfns = 0;
271 
272 	while ((page = balloon_page_pop(&pages))) {
273 		balloon_page_enqueue(&vb->vb_dev_info, page);
274 
275 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
276 		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
277 		if (!virtio_has_feature(vb->vdev,
278 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
279 			adjust_managed_page_count(page, -1);
280 		vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
281 	}
282 
283 	num_allocated_pages = vb->num_pfns;
284 	/* Did we get any? */
285 	if (vb->num_pfns != 0)
286 		tell_host(vb, vb->inflate_vq);
287 	mutex_unlock(&vb->balloon_lock);
288 
289 	return num_allocated_pages;
290 }
291 
release_pages_balloon(struct virtio_balloon * vb,struct list_head * pages)292 static void release_pages_balloon(struct virtio_balloon *vb,
293 				 struct list_head *pages)
294 {
295 	struct page *page, *next;
296 
297 	list_for_each_entry_safe(page, next, pages, lru) {
298 		if (!virtio_has_feature(vb->vdev,
299 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
300 			adjust_managed_page_count(page, 1);
301 		list_del(&page->lru);
302 		put_page(page); /* balloon reference */
303 	}
304 }
305 
leak_balloon(struct virtio_balloon * vb,size_t num)306 static unsigned int leak_balloon(struct virtio_balloon *vb, size_t num)
307 {
308 	unsigned int num_freed_pages;
309 	struct page *page;
310 	struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
311 	LIST_HEAD(pages);
312 
313 	/* We can only do one array worth at a time. */
314 	num = min(num, ARRAY_SIZE(vb->pfns));
315 
316 	mutex_lock(&vb->balloon_lock);
317 	/* We can't release more pages than taken */
318 	num = min(num, (size_t)vb->num_pages);
319 	for (vb->num_pfns = 0; vb->num_pfns < num;
320 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
321 		page = balloon_page_dequeue(vb_dev_info);
322 		if (!page)
323 			break;
324 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
325 		list_add(&page->lru, &pages);
326 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
327 	}
328 
329 	num_freed_pages = vb->num_pfns;
330 	/*
331 	 * Note that if
332 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
333 	 * is true, we *have* to do it in this order
334 	 */
335 	if (vb->num_pfns != 0)
336 		tell_host(vb, vb->deflate_vq);
337 	release_pages_balloon(vb, &pages);
338 	mutex_unlock(&vb->balloon_lock);
339 	return num_freed_pages;
340 }
341 
update_stat(struct virtio_balloon * vb,int idx,u16 tag,u64 val)342 static inline void update_stat(struct virtio_balloon *vb, int idx,
343 			       u16 tag, u64 val)
344 {
345 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
346 	vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
347 	vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
348 }
349 
350 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
351 
352 #ifdef CONFIG_VM_EVENT_COUNTERS
353 /* Return the number of entries filled by vm events */
update_balloon_vm_stats(struct virtio_balloon * vb)354 static inline unsigned int update_balloon_vm_stats(struct virtio_balloon *vb)
355 {
356 	unsigned long events[NR_VM_EVENT_ITEMS];
357 	unsigned int idx = 0;
358 	unsigned int zid;
359 	unsigned long stall = 0;
360 
361 	all_vm_events(events);
362 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
363 		    pages_to_bytes(events[PSWPIN]));
364 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
365 		    pages_to_bytes(events[PSWPOUT]));
366 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
367 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
368 	update_stat(vb, idx++, VIRTIO_BALLOON_S_OOM_KILL, events[OOM_KILL]);
369 
370 	/* sum all the stall events */
371 	for (zid = 0; zid < MAX_NR_ZONES; zid++)
372 		stall += events[ALLOCSTALL_NORMAL - ZONE_NORMAL + zid];
373 
374 	update_stat(vb, idx++, VIRTIO_BALLOON_S_ALLOC_STALL, stall);
375 
376 	update_stat(vb, idx++, VIRTIO_BALLOON_S_ASYNC_SCAN,
377 		    pages_to_bytes(events[PGSCAN_KSWAPD]));
378 	update_stat(vb, idx++, VIRTIO_BALLOON_S_DIRECT_SCAN,
379 		    pages_to_bytes(events[PGSCAN_DIRECT]));
380 	update_stat(vb, idx++, VIRTIO_BALLOON_S_ASYNC_RECLAIM,
381 		    pages_to_bytes(events[PGSTEAL_KSWAPD]));
382 	update_stat(vb, idx++, VIRTIO_BALLOON_S_DIRECT_RECLAIM,
383 		    pages_to_bytes(events[PGSTEAL_DIRECT]));
384 
385 #ifdef CONFIG_HUGETLB_PAGE
386 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC,
387 		    events[HTLB_BUDDY_PGALLOC]);
388 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL,
389 		    events[HTLB_BUDDY_PGALLOC_FAIL]);
390 #endif /* CONFIG_HUGETLB_PAGE */
391 
392 	return idx;
393 }
394 #else /* CONFIG_VM_EVENT_COUNTERS */
update_balloon_vm_stats(struct virtio_balloon * vb)395 static inline unsigned int update_balloon_vm_stats(struct virtio_balloon *vb)
396 {
397 	return 0;
398 }
399 #endif /* CONFIG_VM_EVENT_COUNTERS */
400 
update_balloon_stats(struct virtio_balloon * vb)401 static unsigned int update_balloon_stats(struct virtio_balloon *vb)
402 {
403 	struct sysinfo i;
404 	unsigned int idx;
405 	long available;
406 	unsigned long caches;
407 
408 	idx = update_balloon_vm_stats(vb);
409 
410 	si_meminfo(&i);
411 	available = si_mem_available();
412 	caches = global_node_page_state(NR_FILE_PAGES);
413 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
414 				pages_to_bytes(i.freeram));
415 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
416 				pages_to_bytes(i.totalram));
417 	update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
418 				pages_to_bytes(available));
419 	update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES,
420 				pages_to_bytes(caches));
421 
422 	return idx;
423 }
424 
425 /*
426  * While most virtqueues communicate guest-initiated requests to the hypervisor,
427  * the stats queue operates in reverse.  The driver initializes the virtqueue
428  * with a single buffer.  From that point forward, all conversations consist of
429  * a hypervisor request (a call to this function) which directs us to refill
430  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
431  * we delegate the job to a freezable workqueue that will do the actual work via
432  * stats_handle_request().
433  */
stats_request(struct virtqueue * vq)434 static void stats_request(struct virtqueue *vq)
435 {
436 	struct virtio_balloon *vb = vq->vdev->priv;
437 
438 	spin_lock(&vb->stop_update_lock);
439 	if (!vb->stop_update) {
440 		start_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS);
441 		queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
442 	}
443 	spin_unlock(&vb->stop_update_lock);
444 }
445 
stats_handle_request(struct virtio_balloon * vb)446 static void stats_handle_request(struct virtio_balloon *vb)
447 {
448 	struct virtqueue *vq;
449 	struct scatterlist sg;
450 	unsigned int len, num_stats;
451 
452 	num_stats = update_balloon_stats(vb);
453 
454 	vq = vb->stats_vq;
455 	if (!virtqueue_get_buf(vq, &len))
456 		return;
457 	sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
458 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
459 	virtqueue_kick(vq);
460 }
461 
towards_target(struct virtio_balloon * vb)462 static inline s64 towards_target(struct virtio_balloon *vb)
463 {
464 	s64 target;
465 	u32 num_pages;
466 
467 	/* Legacy balloon config space is LE, unlike all other devices. */
468 	virtio_cread_le(vb->vdev, struct virtio_balloon_config, num_pages,
469 			&num_pages);
470 
471 	/*
472 	 * Aligned up to guest page size to avoid inflating and deflating
473 	 * balloon endlessly.
474 	 */
475 	target = ALIGN(num_pages, VIRTIO_BALLOON_PAGES_PER_PAGE);
476 	return target - vb->num_pages;
477 }
478 
479 /* Gives back @num_to_return blocks of free pages to mm. */
return_free_pages_to_mm(struct virtio_balloon * vb,unsigned long num_to_return)480 static unsigned long return_free_pages_to_mm(struct virtio_balloon *vb,
481 					     unsigned long num_to_return)
482 {
483 	struct page *page;
484 	unsigned long num_returned;
485 
486 	spin_lock_irq(&vb->free_page_list_lock);
487 	for (num_returned = 0; num_returned < num_to_return; num_returned++) {
488 		page = balloon_page_pop(&vb->free_page_list);
489 		if (!page)
490 			break;
491 		__free_pages(page, VIRTIO_BALLOON_HINT_BLOCK_ORDER);
492 	}
493 	vb->num_free_page_blocks -= num_returned;
494 	spin_unlock_irq(&vb->free_page_list_lock);
495 
496 	return num_returned;
497 }
498 
virtio_balloon_queue_free_page_work(struct virtio_balloon * vb)499 static void virtio_balloon_queue_free_page_work(struct virtio_balloon *vb)
500 {
501 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
502 		return;
503 
504 	/* No need to queue the work if the bit was already set. */
505 	if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
506 			     &vb->config_read_bitmap))
507 		return;
508 
509 	queue_work(vb->balloon_wq, &vb->report_free_page_work);
510 }
511 
start_update_balloon_size(struct virtio_balloon * vb)512 static void start_update_balloon_size(struct virtio_balloon *vb)
513 {
514 	start_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST);
515 	queue_work(system_freezable_wq, &vb->update_balloon_size_work);
516 }
517 
virtballoon_changed(struct virtio_device * vdev)518 static void virtballoon_changed(struct virtio_device *vdev)
519 {
520 	struct virtio_balloon *vb = vdev->priv;
521 	unsigned long flags;
522 
523 	spin_lock_irqsave(&vb->stop_update_lock, flags);
524 	if (!vb->stop_update) {
525 		start_update_balloon_size(vb);
526 		virtio_balloon_queue_free_page_work(vb);
527 	}
528 	spin_unlock_irqrestore(&vb->stop_update_lock, flags);
529 }
530 
update_balloon_size(struct virtio_balloon * vb)531 static void update_balloon_size(struct virtio_balloon *vb)
532 {
533 	u32 actual = vb->num_pages;
534 
535 	/* Legacy balloon config space is LE, unlike all other devices. */
536 	virtio_cwrite_le(vb->vdev, struct virtio_balloon_config, actual,
537 			 &actual);
538 }
539 
update_balloon_stats_func(struct work_struct * work)540 static void update_balloon_stats_func(struct work_struct *work)
541 {
542 	struct virtio_balloon *vb;
543 
544 	vb = container_of(work, struct virtio_balloon,
545 			  update_balloon_stats_work);
546 
547 	process_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS);
548 	stats_handle_request(vb);
549 	finish_wakeup_event(vb);
550 }
551 
update_balloon_size_func(struct work_struct * work)552 static void update_balloon_size_func(struct work_struct *work)
553 {
554 	struct virtio_balloon *vb;
555 	s64 diff;
556 
557 	vb = container_of(work, struct virtio_balloon,
558 			  update_balloon_size_work);
559 
560 	process_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST);
561 
562 	diff = towards_target(vb);
563 
564 	if (diff) {
565 		if (diff > 0)
566 			diff -= fill_balloon(vb, diff);
567 		else
568 			diff += leak_balloon(vb, -diff);
569 		update_balloon_size(vb);
570 	}
571 
572 	if (diff)
573 		queue_work(system_freezable_wq, work);
574 	else
575 		finish_wakeup_event(vb);
576 }
577 
init_vqs(struct virtio_balloon * vb)578 static int init_vqs(struct virtio_balloon *vb)
579 {
580 	struct virtqueue_info vqs_info[VIRTIO_BALLOON_VQ_MAX] = {};
581 	struct virtqueue *vqs[VIRTIO_BALLOON_VQ_MAX];
582 	int err;
583 
584 	/*
585 	 * Inflateq and deflateq are used unconditionally. The names[]
586 	 * will be NULL if the related feature is not enabled, which will
587 	 * cause no allocation for the corresponding virtqueue in find_vqs.
588 	 */
589 	vqs_info[VIRTIO_BALLOON_VQ_INFLATE].callback = balloon_ack;
590 	vqs_info[VIRTIO_BALLOON_VQ_INFLATE].name = "inflate";
591 	vqs_info[VIRTIO_BALLOON_VQ_DEFLATE].callback = balloon_ack;
592 	vqs_info[VIRTIO_BALLOON_VQ_DEFLATE].name = "deflate";
593 
594 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
595 		vqs_info[VIRTIO_BALLOON_VQ_STATS].name = "stats";
596 		vqs_info[VIRTIO_BALLOON_VQ_STATS].callback = stats_request;
597 	}
598 
599 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
600 		vqs_info[VIRTIO_BALLOON_VQ_FREE_PAGE].name = "free_page_vq";
601 
602 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) {
603 		vqs_info[VIRTIO_BALLOON_VQ_REPORTING].name = "reporting_vq";
604 		vqs_info[VIRTIO_BALLOON_VQ_REPORTING].callback = balloon_ack;
605 	}
606 
607 	err = virtio_find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX, vqs,
608 			      vqs_info, NULL);
609 	if (err)
610 		return err;
611 
612 	vb->inflate_vq = vqs[VIRTIO_BALLOON_VQ_INFLATE];
613 	vb->deflate_vq = vqs[VIRTIO_BALLOON_VQ_DEFLATE];
614 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
615 		struct scatterlist sg;
616 		unsigned int num_stats;
617 		vb->stats_vq = vqs[VIRTIO_BALLOON_VQ_STATS];
618 
619 		/*
620 		 * Prime this virtqueue with one buffer so the hypervisor can
621 		 * use it to signal us later (it can't be broken yet!).
622 		 */
623 		num_stats = update_balloon_stats(vb);
624 
625 		sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
626 		err = virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb,
627 					   GFP_KERNEL);
628 		if (err) {
629 			dev_warn(&vb->vdev->dev, "%s: add stat_vq failed\n",
630 				 __func__);
631 			return err;
632 		}
633 		virtqueue_kick(vb->stats_vq);
634 	}
635 
636 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
637 		vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE];
638 
639 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING))
640 		vb->reporting_vq = vqs[VIRTIO_BALLOON_VQ_REPORTING];
641 
642 	return 0;
643 }
644 
virtio_balloon_cmd_id_received(struct virtio_balloon * vb)645 static u32 virtio_balloon_cmd_id_received(struct virtio_balloon *vb)
646 {
647 	if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
648 			       &vb->config_read_bitmap)) {
649 		/* Legacy balloon config space is LE, unlike all other devices. */
650 		virtio_cread_le(vb->vdev, struct virtio_balloon_config,
651 				free_page_hint_cmd_id,
652 				&vb->cmd_id_received_cache);
653 	}
654 
655 	return vb->cmd_id_received_cache;
656 }
657 
send_cmd_id_start(struct virtio_balloon * vb)658 static int send_cmd_id_start(struct virtio_balloon *vb)
659 {
660 	struct scatterlist sg;
661 	struct virtqueue *vq = vb->free_page_vq;
662 	int err, unused;
663 
664 	/* Detach all the used buffers from the vq */
665 	while (virtqueue_get_buf(vq, &unused))
666 		;
667 
668 	vb->cmd_id_active = cpu_to_virtio32(vb->vdev,
669 					virtio_balloon_cmd_id_received(vb));
670 	sg_init_one(&sg, &vb->cmd_id_active, sizeof(vb->cmd_id_active));
671 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_active, GFP_KERNEL);
672 	if (!err)
673 		virtqueue_kick(vq);
674 	return err;
675 }
676 
send_cmd_id_stop(struct virtio_balloon * vb)677 static int send_cmd_id_stop(struct virtio_balloon *vb)
678 {
679 	struct scatterlist sg;
680 	struct virtqueue *vq = vb->free_page_vq;
681 	int err, unused;
682 
683 	/* Detach all the used buffers from the vq */
684 	while (virtqueue_get_buf(vq, &unused))
685 		;
686 
687 	sg_init_one(&sg, &vb->cmd_id_stop, sizeof(vb->cmd_id_stop));
688 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_stop, GFP_KERNEL);
689 	if (!err)
690 		virtqueue_kick(vq);
691 	return err;
692 }
693 
get_free_page_and_send(struct virtio_balloon * vb)694 static int get_free_page_and_send(struct virtio_balloon *vb)
695 {
696 	struct virtqueue *vq = vb->free_page_vq;
697 	struct page *page;
698 	struct scatterlist sg;
699 	int err, unused;
700 	void *p;
701 
702 	/* Detach all the used buffers from the vq */
703 	while (virtqueue_get_buf(vq, &unused))
704 		;
705 
706 	page = alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG,
707 			   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
708 	/*
709 	 * When the allocation returns NULL, it indicates that we have got all
710 	 * the possible free pages, so return -EINTR to stop.
711 	 */
712 	if (!page)
713 		return -EINTR;
714 
715 	p = page_address(page);
716 	sg_init_one(&sg, p, VIRTIO_BALLOON_HINT_BLOCK_BYTES);
717 	/* There is always 1 entry reserved for the cmd id to use. */
718 	if (vq->num_free > 1) {
719 		err = virtqueue_add_inbuf(vq, &sg, 1, p, GFP_KERNEL);
720 		if (unlikely(err)) {
721 			__free_pages(page, VIRTIO_BALLOON_HINT_BLOCK_ORDER);
722 			return err;
723 		}
724 		virtqueue_kick(vq);
725 		spin_lock_irq(&vb->free_page_list_lock);
726 		balloon_page_push(&vb->free_page_list, page);
727 		vb->num_free_page_blocks++;
728 		spin_unlock_irq(&vb->free_page_list_lock);
729 	} else {
730 		/*
731 		 * The vq has no available entry to add this page block, so
732 		 * just free it.
733 		 */
734 		__free_pages(page, VIRTIO_BALLOON_HINT_BLOCK_ORDER);
735 	}
736 
737 	return 0;
738 }
739 
send_free_pages(struct virtio_balloon * vb)740 static int send_free_pages(struct virtio_balloon *vb)
741 {
742 	int err;
743 	u32 cmd_id_active;
744 
745 	while (1) {
746 		/*
747 		 * If a stop id or a new cmd id was just received from host,
748 		 * stop the reporting.
749 		 */
750 		cmd_id_active = virtio32_to_cpu(vb->vdev, vb->cmd_id_active);
751 		if (unlikely(cmd_id_active !=
752 			     virtio_balloon_cmd_id_received(vb)))
753 			break;
754 
755 		/*
756 		 * The free page blocks are allocated and sent to host one by
757 		 * one.
758 		 */
759 		err = get_free_page_and_send(vb);
760 		if (err == -EINTR)
761 			break;
762 		else if (unlikely(err))
763 			return err;
764 	}
765 
766 	return 0;
767 }
768 
virtio_balloon_report_free_page(struct virtio_balloon * vb)769 static void virtio_balloon_report_free_page(struct virtio_balloon *vb)
770 {
771 	int err;
772 	struct device *dev = &vb->vdev->dev;
773 
774 	/* Start by sending the received cmd id to host with an outbuf. */
775 	err = send_cmd_id_start(vb);
776 	if (unlikely(err))
777 		dev_err(dev, "Failed to send a start id, err = %d\n", err);
778 
779 	err = send_free_pages(vb);
780 	if (unlikely(err))
781 		dev_err(dev, "Failed to send a free page, err = %d\n", err);
782 
783 	/* End by sending a stop id to host with an outbuf. */
784 	err = send_cmd_id_stop(vb);
785 	if (unlikely(err))
786 		dev_err(dev, "Failed to send a stop id, err = %d\n", err);
787 }
788 
report_free_page_func(struct work_struct * work)789 static void report_free_page_func(struct work_struct *work)
790 {
791 	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
792 						 report_free_page_work);
793 	u32 cmd_id_received;
794 
795 	cmd_id_received = virtio_balloon_cmd_id_received(vb);
796 	if (cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
797 		/* Pass ULONG_MAX to give back all the free pages */
798 		return_free_pages_to_mm(vb, ULONG_MAX);
799 	} else if (cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
800 		   cmd_id_received !=
801 		   virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) {
802 		virtio_balloon_report_free_page(vb);
803 	}
804 }
805 
806 #ifdef CONFIG_BALLOON_COMPACTION
807 /*
808  * virtballoon_migratepage - perform the balloon page migration on behalf of
809  *			     a compaction thread.     (called under page lock)
810  * @vb_dev_info: the balloon device
811  * @newpage: page that will replace the isolated page after migration finishes.
812  * @page   : the isolated (old) page that is about to be migrated to newpage.
813  * @mode   : compaction mode -- not used for balloon page migration.
814  *
815  * After a ballooned page gets isolated by compaction procedures, this is the
816  * function that performs the page migration on behalf of a compaction thread
817  * The page migration for virtio balloon is done in a simple swap fashion which
818  * follows these two macro steps:
819  *  1) insert newpage into vb->pages list and update the host about it;
820  *  2) update the host about the old page removed from vb->pages list;
821  *
822  * This function preforms the balloon page migration task.
823  * Called through movable_operations->migrate_page
824  */
virtballoon_migratepage(struct balloon_dev_info * vb_dev_info,struct page * newpage,struct page * page,enum migrate_mode mode)825 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
826 		struct page *newpage, struct page *page, enum migrate_mode mode)
827 {
828 	struct virtio_balloon *vb = container_of(vb_dev_info,
829 			struct virtio_balloon, vb_dev_info);
830 	unsigned long flags;
831 
832 	/*
833 	 * In order to avoid lock contention while migrating pages concurrently
834 	 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
835 	 * this turn, as it is easier to retry the page migration later.
836 	 * This also prevents fill_balloon() getting stuck into a mutex
837 	 * recursion in the case it ends up triggering memory compaction
838 	 * while it is attempting to inflate the ballon.
839 	 */
840 	if (!mutex_trylock(&vb->balloon_lock))
841 		return -EAGAIN;
842 
843 	get_page(newpage); /* balloon reference */
844 
845 	/*
846 	  * When we migrate a page to a different zone and adjusted the
847 	  * managed page count when inflating, we have to fixup the count of
848 	  * both involved zones.
849 	  */
850 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM) &&
851 	    page_zone(page) != page_zone(newpage)) {
852 		adjust_managed_page_count(page, 1);
853 		adjust_managed_page_count(newpage, -1);
854 	}
855 
856 	/* balloon's page migration 1st step  -- inflate "newpage" */
857 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
858 	balloon_page_insert(vb_dev_info, newpage);
859 	vb_dev_info->isolated_pages--;
860 	__count_vm_event(BALLOON_MIGRATE);
861 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
862 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
863 	set_page_pfns(vb, vb->pfns, newpage);
864 	tell_host(vb, vb->inflate_vq);
865 
866 	/* balloon's page migration 2nd step -- deflate "page" */
867 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
868 	set_page_pfns(vb, vb->pfns, page);
869 	tell_host(vb, vb->deflate_vq);
870 
871 	mutex_unlock(&vb->balloon_lock);
872 
873 	balloon_page_finalize(page);
874 	put_page(page); /* balloon reference */
875 
876 	return 0;
877 }
878 #endif /* CONFIG_BALLOON_COMPACTION */
879 
shrink_free_pages(struct virtio_balloon * vb,unsigned long pages_to_free)880 static unsigned long shrink_free_pages(struct virtio_balloon *vb,
881 				       unsigned long pages_to_free)
882 {
883 	unsigned long blocks_to_free, blocks_freed;
884 
885 	pages_to_free = round_up(pages_to_free,
886 				 VIRTIO_BALLOON_HINT_BLOCK_PAGES);
887 	blocks_to_free = pages_to_free / VIRTIO_BALLOON_HINT_BLOCK_PAGES;
888 	blocks_freed = return_free_pages_to_mm(vb, blocks_to_free);
889 
890 	return blocks_freed * VIRTIO_BALLOON_HINT_BLOCK_PAGES;
891 }
892 
virtio_balloon_shrinker_scan(struct shrinker * shrinker,struct shrink_control * sc)893 static unsigned long virtio_balloon_shrinker_scan(struct shrinker *shrinker,
894 						  struct shrink_control *sc)
895 {
896 	struct virtio_balloon *vb = shrinker->private_data;
897 
898 	return shrink_free_pages(vb, sc->nr_to_scan);
899 }
900 
virtio_balloon_shrinker_count(struct shrinker * shrinker,struct shrink_control * sc)901 static unsigned long virtio_balloon_shrinker_count(struct shrinker *shrinker,
902 						   struct shrink_control *sc)
903 {
904 	struct virtio_balloon *vb = shrinker->private_data;
905 
906 	return vb->num_free_page_blocks * VIRTIO_BALLOON_HINT_BLOCK_PAGES;
907 }
908 
virtio_balloon_oom_notify(struct notifier_block * nb,unsigned long dummy,void * parm)909 static int virtio_balloon_oom_notify(struct notifier_block *nb,
910 				     unsigned long dummy, void *parm)
911 {
912 	struct virtio_balloon *vb = container_of(nb,
913 						 struct virtio_balloon, oom_nb);
914 	unsigned long *freed = parm;
915 
916 	*freed += leak_balloon(vb, VIRTIO_BALLOON_OOM_NR_PAGES) /
917 		  VIRTIO_BALLOON_PAGES_PER_PAGE;
918 	update_balloon_size(vb);
919 
920 	return NOTIFY_OK;
921 }
922 
virtio_balloon_unregister_shrinker(struct virtio_balloon * vb)923 static void virtio_balloon_unregister_shrinker(struct virtio_balloon *vb)
924 {
925 	shrinker_free(vb->shrinker);
926 }
927 
virtio_balloon_register_shrinker(struct virtio_balloon * vb)928 static int virtio_balloon_register_shrinker(struct virtio_balloon *vb)
929 {
930 	vb->shrinker = shrinker_alloc(0, "virtio-balloon");
931 	if (!vb->shrinker)
932 		return -ENOMEM;
933 
934 	vb->shrinker->scan_objects = virtio_balloon_shrinker_scan;
935 	vb->shrinker->count_objects = virtio_balloon_shrinker_count;
936 	vb->shrinker->private_data = vb;
937 
938 	shrinker_register(vb->shrinker);
939 
940 	return 0;
941 }
942 
virtballoon_probe(struct virtio_device * vdev)943 static int virtballoon_probe(struct virtio_device *vdev)
944 {
945 	struct virtio_balloon *vb;
946 	int err;
947 
948 	if (!vdev->config->get) {
949 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
950 			__func__);
951 		return -EINVAL;
952 	}
953 
954 	vdev->priv = vb = kzalloc(sizeof(*vb), GFP_KERNEL);
955 	if (!vb) {
956 		err = -ENOMEM;
957 		goto out;
958 	}
959 
960 	INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
961 	INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
962 	spin_lock_init(&vb->stop_update_lock);
963 	mutex_init(&vb->balloon_lock);
964 	init_waitqueue_head(&vb->acked);
965 	vb->vdev = vdev;
966 
967 	balloon_devinfo_init(&vb->vb_dev_info);
968 
969 	err = init_vqs(vb);
970 	if (err)
971 		goto out_free_vb;
972 
973 #ifdef CONFIG_BALLOON_COMPACTION
974 	vb->vb_dev_info.migratepage = virtballoon_migratepage;
975 #endif
976 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
977 		/*
978 		 * There is always one entry reserved for cmd id, so the ring
979 		 * size needs to be at least two to report free page hints.
980 		 */
981 		if (virtqueue_get_vring_size(vb->free_page_vq) < 2) {
982 			err = -ENOSPC;
983 			goto out_del_vqs;
984 		}
985 		vb->balloon_wq = alloc_workqueue("balloon-wq",
986 					WQ_FREEZABLE | WQ_CPU_INTENSIVE, 0);
987 		if (!vb->balloon_wq) {
988 			err = -ENOMEM;
989 			goto out_del_vqs;
990 		}
991 		INIT_WORK(&vb->report_free_page_work, report_free_page_func);
992 		vb->cmd_id_received_cache = VIRTIO_BALLOON_CMD_ID_STOP;
993 		vb->cmd_id_active = cpu_to_virtio32(vb->vdev,
994 						  VIRTIO_BALLOON_CMD_ID_STOP);
995 		vb->cmd_id_stop = cpu_to_virtio32(vb->vdev,
996 						  VIRTIO_BALLOON_CMD_ID_STOP);
997 		spin_lock_init(&vb->free_page_list_lock);
998 		INIT_LIST_HEAD(&vb->free_page_list);
999 		/*
1000 		 * We're allowed to reuse any free pages, even if they are
1001 		 * still to be processed by the host.
1002 		 */
1003 		err = virtio_balloon_register_shrinker(vb);
1004 		if (err)
1005 			goto out_del_balloon_wq;
1006 	}
1007 
1008 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) {
1009 		vb->oom_nb.notifier_call = virtio_balloon_oom_notify;
1010 		vb->oom_nb.priority = VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY;
1011 		err = register_oom_notifier(&vb->oom_nb);
1012 		if (err < 0)
1013 			goto out_unregister_shrinker;
1014 	}
1015 
1016 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) {
1017 		/* Start with poison val of 0 representing general init */
1018 		__u32 poison_val = 0;
1019 
1020 		/*
1021 		 * Let the hypervisor know that we are expecting a
1022 		 * specific value to be written back in balloon pages.
1023 		 *
1024 		 * If the PAGE_POISON value was larger than a byte we would
1025 		 * need to byte swap poison_val here to guarantee it is
1026 		 * little-endian. However for now it is a single byte so we
1027 		 * can pass it as-is.
1028 		 */
1029 		if (!want_init_on_free())
1030 			memset(&poison_val, PAGE_POISON, sizeof(poison_val));
1031 
1032 		virtio_cwrite_le(vb->vdev, struct virtio_balloon_config,
1033 				 poison_val, &poison_val);
1034 	}
1035 
1036 	vb->pr_dev_info.report = virtballoon_free_page_report;
1037 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) {
1038 		unsigned int capacity;
1039 
1040 		capacity = virtqueue_get_vring_size(vb->reporting_vq);
1041 		if (capacity < PAGE_REPORTING_CAPACITY) {
1042 			err = -ENOSPC;
1043 			goto out_unregister_oom;
1044 		}
1045 
1046 		/*
1047 		 * The default page reporting order is @pageblock_order, which
1048 		 * corresponds to 512MB in size on ARM64 when 64KB base page
1049 		 * size is used. The page reporting won't be triggered if the
1050 		 * freeing page can't come up with a free area like that huge.
1051 		 * So we specify the page reporting order to 5, corresponding
1052 		 * to 2MB. It helps to avoid THP splitting if 4KB base page
1053 		 * size is used by host.
1054 		 *
1055 		 * Ideally, the page reporting order is selected based on the
1056 		 * host's base page size. However, it needs more work to report
1057 		 * that value. The hard-coded order would be fine currently.
1058 		 */
1059 #if defined(CONFIG_ARM64) && defined(CONFIG_ARM64_64K_PAGES)
1060 		vb->pr_dev_info.order = 5;
1061 #endif
1062 
1063 		err = page_reporting_register(&vb->pr_dev_info);
1064 		if (err)
1065 			goto out_unregister_oom;
1066 	}
1067 
1068 	spin_lock_init(&vb->wakeup_lock);
1069 
1070 	/*
1071 	 * The virtio balloon itself can't wake up the device, but it is
1072 	 * responsible for processing wakeup events passed up from the transport
1073 	 * layer. Wakeup sources don't support nesting/chaining calls, so we use
1074 	 * our own wakeup source to ensure wakeup events are properly handled
1075 	 * without trampling on the transport layer's wakeup source.
1076 	 */
1077 	device_set_wakeup_capable(&vb->vdev->dev, true);
1078 
1079 	virtio_device_ready(vdev);
1080 
1081 	if (towards_target(vb))
1082 		virtballoon_changed(vdev);
1083 	return 0;
1084 
1085 out_unregister_oom:
1086 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
1087 		unregister_oom_notifier(&vb->oom_nb);
1088 out_unregister_shrinker:
1089 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1090 		virtio_balloon_unregister_shrinker(vb);
1091 out_del_balloon_wq:
1092 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1093 		destroy_workqueue(vb->balloon_wq);
1094 out_del_vqs:
1095 	vdev->config->del_vqs(vdev);
1096 out_free_vb:
1097 	kfree(vb);
1098 out:
1099 	return err;
1100 }
1101 
remove_common(struct virtio_balloon * vb)1102 static void remove_common(struct virtio_balloon *vb)
1103 {
1104 	/* There might be pages left in the balloon: free them. */
1105 	while (vb->num_pages)
1106 		leak_balloon(vb, vb->num_pages);
1107 	update_balloon_size(vb);
1108 
1109 	/* There might be free pages that are being reported: release them. */
1110 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1111 		return_free_pages_to_mm(vb, ULONG_MAX);
1112 
1113 	/* Now we reset the device so we can clean up the queues. */
1114 	virtio_reset_device(vb->vdev);
1115 
1116 	vb->vdev->config->del_vqs(vb->vdev);
1117 }
1118 
virtballoon_remove(struct virtio_device * vdev)1119 static void virtballoon_remove(struct virtio_device *vdev)
1120 {
1121 	struct virtio_balloon *vb = vdev->priv;
1122 
1123 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING))
1124 		page_reporting_unregister(&vb->pr_dev_info);
1125 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
1126 		unregister_oom_notifier(&vb->oom_nb);
1127 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1128 		virtio_balloon_unregister_shrinker(vb);
1129 	spin_lock_irq(&vb->stop_update_lock);
1130 	vb->stop_update = true;
1131 	spin_unlock_irq(&vb->stop_update_lock);
1132 	cancel_work_sync(&vb->update_balloon_size_work);
1133 	cancel_work_sync(&vb->update_balloon_stats_work);
1134 
1135 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
1136 		cancel_work_sync(&vb->report_free_page_work);
1137 		destroy_workqueue(vb->balloon_wq);
1138 	}
1139 
1140 	remove_common(vb);
1141 	kfree(vb);
1142 }
1143 
1144 #ifdef CONFIG_PM_SLEEP
virtballoon_freeze(struct virtio_device * vdev)1145 static int virtballoon_freeze(struct virtio_device *vdev)
1146 {
1147 	struct virtio_balloon *vb = vdev->priv;
1148 
1149 	/*
1150 	 * The workqueue is already frozen by the PM core before this
1151 	 * function is called.
1152 	 */
1153 	remove_common(vb);
1154 	return 0;
1155 }
1156 
virtballoon_restore(struct virtio_device * vdev)1157 static int virtballoon_restore(struct virtio_device *vdev)
1158 {
1159 	struct virtio_balloon *vb = vdev->priv;
1160 	int ret;
1161 
1162 	ret = init_vqs(vdev->priv);
1163 	if (ret)
1164 		return ret;
1165 
1166 	virtio_device_ready(vdev);
1167 
1168 	if (towards_target(vb))
1169 		virtballoon_changed(vdev);
1170 	update_balloon_size(vb);
1171 	return 0;
1172 }
1173 #endif
1174 
virtballoon_validate(struct virtio_device * vdev)1175 static int virtballoon_validate(struct virtio_device *vdev)
1176 {
1177 	/*
1178 	 * Inform the hypervisor that our pages are poisoned or
1179 	 * initialized. If we cannot do that then we should disable
1180 	 * page reporting as it could potentially change the contents
1181 	 * of our free pages.
1182 	 */
1183 	if (!want_init_on_free() && !page_poisoning_enabled_static())
1184 		__virtio_clear_bit(vdev, VIRTIO_BALLOON_F_PAGE_POISON);
1185 	else if (!virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON))
1186 		__virtio_clear_bit(vdev, VIRTIO_BALLOON_F_REPORTING);
1187 
1188 	__virtio_clear_bit(vdev, VIRTIO_F_ACCESS_PLATFORM);
1189 	return 0;
1190 }
1191 
1192 static unsigned int features[] = {
1193 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
1194 	VIRTIO_BALLOON_F_STATS_VQ,
1195 	VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
1196 	VIRTIO_BALLOON_F_FREE_PAGE_HINT,
1197 	VIRTIO_BALLOON_F_PAGE_POISON,
1198 	VIRTIO_BALLOON_F_REPORTING,
1199 };
1200 
1201 static struct virtio_driver virtio_balloon_driver = {
1202 	.feature_table = features,
1203 	.feature_table_size = ARRAY_SIZE(features),
1204 	.driver.name =	KBUILD_MODNAME,
1205 	.id_table =	id_table,
1206 	.validate =	virtballoon_validate,
1207 	.probe =	virtballoon_probe,
1208 	.remove =	virtballoon_remove,
1209 	.config_changed = virtballoon_changed,
1210 #ifdef CONFIG_PM_SLEEP
1211 	.freeze	=	virtballoon_freeze,
1212 	.restore =	virtballoon_restore,
1213 #endif
1214 };
1215 
1216 module_virtio_driver(virtio_balloon_driver);
1217 MODULE_DEVICE_TABLE(virtio, id_table);
1218 MODULE_DESCRIPTION("Virtio balloon driver");
1219 MODULE_LICENSE("GPL");
1220