• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include <linux/kstrtox.h>
22 
23 /*
24  * Balloon device works in 4K page units.  So each page is pointed to by
25  * multiple balloon pages.  All memory counters in this driver are in balloon
26  * page units.
27  */
28 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned int)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
29 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
30 /* Maximum number of (4k) pages to deflate on OOM notifications. */
31 #define VIRTIO_BALLOON_OOM_NR_PAGES 256
32 #define VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY 80
33 
34 #define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \
35 					     __GFP_NOMEMALLOC)
36 /* The order of free page blocks to report to host */
37 #define VIRTIO_BALLOON_HINT_BLOCK_ORDER MAX_PAGE_ORDER
38 /* The size of a free page block in bytes */
39 #define VIRTIO_BALLOON_HINT_BLOCK_BYTES \
40 	(1 << (VIRTIO_BALLOON_HINT_BLOCK_ORDER + PAGE_SHIFT))
41 #define VIRTIO_BALLOON_HINT_BLOCK_PAGES (1 << VIRTIO_BALLOON_HINT_BLOCK_ORDER)
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_REPORTING,
49 	VIRTIO_BALLOON_VQ_MAX
50 };
51 
52 enum virtio_balloon_config_read {
53 	VIRTIO_BALLOON_CONFIG_READ_CMD_ID = 0,
54 };
55 
56 struct virtio_balloon {
57 	struct virtio_device *vdev;
58 	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq;
59 
60 	/* Balloon's own wq for cpu-intensive work items */
61 	struct workqueue_struct *balloon_wq;
62 	/* The free page reporting work item submitted to the balloon wq */
63 	struct work_struct report_free_page_work;
64 
65 	/* The balloon servicing is delegated to a freezable workqueue. */
66 	struct work_struct update_balloon_stats_work;
67 	struct work_struct update_balloon_size_work;
68 
69 	/* Prevent updating balloon when it is being canceled. */
70 	spinlock_t stop_update_lock;
71 	bool stop_update;
72 	/* Bitmap to indicate if reading the related config fields are needed */
73 	unsigned long config_read_bitmap;
74 
75 	/* The list of allocated free pages, waiting to be given back to mm */
76 	struct list_head free_page_list;
77 	spinlock_t free_page_list_lock;
78 	/* The number of free page blocks on the above list */
79 	unsigned long num_free_page_blocks;
80 	/*
81 	 * The cmd id received from host.
82 	 * Read it via virtio_balloon_cmd_id_received to get the latest value
83 	 * sent from host.
84 	 */
85 	u32 cmd_id_received_cache;
86 	/* The cmd id that is actively in use */
87 	__virtio32 cmd_id_active;
88 	/* Buffer to store the stop sign */
89 	__virtio32 cmd_id_stop;
90 
91 	/* Waiting for host to ack the pages we released. */
92 	wait_queue_head_t acked;
93 
94 	/* Number of balloon pages we've told the Host we're not using. */
95 	unsigned int num_pages;
96 	/*
97 	 * The pages we've told the Host we're not using are enqueued
98 	 * at vb_dev_info->pages list.
99 	 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
100 	 * to num_pages above.
101 	 */
102 	struct balloon_dev_info vb_dev_info;
103 
104 	/* Synchronize access/update to this struct virtio_balloon elements */
105 	struct mutex balloon_lock;
106 
107 	/* The array of pfns we tell the Host about. */
108 	unsigned int num_pfns;
109 	__virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
110 
111 	/* Memory statistics */
112 	struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
113 
114 	/* Shrinker to return free pages - VIRTIO_BALLOON_F_FREE_PAGE_HINT */
115 	struct shrinker *shrinker;
116 
117 	/* OOM notifier to deflate on OOM - VIRTIO_BALLOON_F_DEFLATE_ON_OOM */
118 	struct notifier_block oom_nb;
119 
120 	/* Free page reporting device */
121 	struct virtqueue *reporting_vq;
122 	struct page_reporting_dev_info pr_dev_info;
123 
124 	/* State for keeping the wakeup_source active while adjusting the balloon */
125 	spinlock_t wakeup_lock;
126 	bool processing_wakeup_event;
127 	u32 wakeup_signal_mask;
128 
129 	bool bail_on_out_of_puff;
130 };
131 
132 #define VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST (1 << 0)
133 #define VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS (1 << 1)
134 
135 static const struct virtio_device_id id_table[] = {
136 	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
137 	{ 0 },
138 };
139 
page_to_balloon_pfn(struct page * page)140 static u32 page_to_balloon_pfn(struct page *page)
141 {
142 	unsigned long pfn = page_to_pfn(page);
143 
144 	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
145 	/* Convert pfn from Linux page size to balloon page size. */
146 	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
147 }
148 
start_wakeup_event(struct virtio_balloon * vb,u32 mask)149 static void start_wakeup_event(struct virtio_balloon *vb, u32 mask)
150 {
151 	unsigned long flags;
152 
153 	spin_lock_irqsave(&vb->wakeup_lock, flags);
154 	vb->wakeup_signal_mask |= mask;
155 	if (!vb->processing_wakeup_event) {
156 		vb->processing_wakeup_event = true;
157 		pm_stay_awake(&vb->vdev->dev);
158 	}
159 	spin_unlock_irqrestore(&vb->wakeup_lock, flags);
160 }
161 
process_wakeup_event(struct virtio_balloon * vb,u32 mask)162 static void process_wakeup_event(struct virtio_balloon *vb, u32 mask)
163 {
164 	spin_lock_irq(&vb->wakeup_lock);
165 	vb->wakeup_signal_mask &= ~mask;
166 	spin_unlock_irq(&vb->wakeup_lock);
167 }
168 
finish_wakeup_event(struct virtio_balloon * vb)169 static void finish_wakeup_event(struct virtio_balloon *vb)
170 {
171 	spin_lock_irq(&vb->wakeup_lock);
172 	if (!vb->wakeup_signal_mask && vb->processing_wakeup_event) {
173 		vb->processing_wakeup_event = false;
174 		pm_relax(&vb->vdev->dev);
175 	}
176 	spin_unlock_irq(&vb->wakeup_lock);
177 }
178 
balloon_ack(struct virtqueue * vq)179 static void balloon_ack(struct virtqueue *vq)
180 {
181 	struct virtio_balloon *vb = vq->vdev->priv;
182 
183 	wake_up(&vb->acked);
184 }
185 
tell_host(struct virtio_balloon * vb,struct virtqueue * vq)186 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
187 {
188 	struct scatterlist sg;
189 	unsigned int len;
190 
191 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
192 
193 	post_page_relinquish_tlb_inv();
194 
195 	/* We should always be able to add one buffer to an empty queue. */
196 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
197 	virtqueue_kick(vq);
198 
199 	/* When host has read buffer, this completes via balloon_ack */
200 	wait_event(vb->acked, virtqueue_get_buf(vq, &len));
201 
202 }
203 
virtballoon_free_page_report(struct page_reporting_dev_info * pr_dev_info,struct scatterlist * sgl,unsigned int nents)204 static int virtballoon_free_page_report(struct page_reporting_dev_info *pr_dev_info,
205 				   struct scatterlist *sgl, unsigned int nents)
206 {
207 	struct virtio_balloon *vb =
208 		container_of(pr_dev_info, struct virtio_balloon, pr_dev_info);
209 	struct virtqueue *vq = vb->reporting_vq;
210 	unsigned int unused, err;
211 	struct scatterlist *sg;
212 
213 	for (sg = sgl; sg != NULL; sg = sg_next(sg))
214 		page_relinquish(sg_page(sg), sg->length >> PAGE_SHIFT);
215 
216 	/* We should always be able to add these buffers to an empty queue. */
217 	err = virtqueue_add_inbuf(vq, sgl, nents, vb, GFP_NOWAIT | __GFP_NOWARN);
218 
219 	/*
220 	 * In the extremely unlikely case that something has occurred and we
221 	 * are able to trigger an error we will simply display a warning
222 	 * and exit without actually processing the pages.
223 	 */
224 	if (WARN_ON_ONCE(err))
225 		return err;
226 
227 	post_page_relinquish_tlb_inv();
228 
229 	virtqueue_kick(vq);
230 
231 	/* When host has read buffer, this completes via balloon_ack */
232 	wait_event(vb->acked, virtqueue_get_buf(vq, &unused));
233 
234 	return 0;
235 }
236 
set_page_pfns(struct virtio_balloon * vb,__virtio32 pfns[],struct page * page)237 static void set_page_pfns(struct virtio_balloon *vb,
238 			  __virtio32 pfns[], struct page *page)
239 {
240 	unsigned int i;
241 
242 	BUILD_BUG_ON(VIRTIO_BALLOON_PAGES_PER_PAGE > VIRTIO_BALLOON_ARRAY_PFNS_MAX);
243 
244 	/*
245 	 * Set balloon pfns pointing at this page.
246 	 * Note that the first pfn points at start of the page.
247 	 */
248 	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
249 		pfns[i] = cpu_to_virtio32(vb->vdev,
250 					  page_to_balloon_pfn(page) + i);
251 }
252 
fill_balloon(struct virtio_balloon * vb,size_t num,bool * out_of_puff)253 static unsigned int fill_balloon(struct virtio_balloon *vb, size_t num,
254 				 bool *out_of_puff)
255 {
256 	unsigned int num_allocated_pages;
257 	unsigned int num_pfns;
258 	struct page *page;
259 	LIST_HEAD(pages);
260 
261 	/* We can only do one array worth at a time. */
262 	num = min(num, ARRAY_SIZE(vb->pfns));
263 
264 	for (num_pfns = 0; num_pfns < num;
265 	     num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
266 		struct page *page = balloon_page_alloc();
267 
268 		if (!page) {
269 			dev_info_ratelimited(&vb->vdev->dev,
270 					     "Out of puff! Can't get %u pages\n",
271 					     VIRTIO_BALLOON_PAGES_PER_PAGE);
272 			/* Sleep for at least 1/5 of a second before retry. */
273 			msleep(200);
274 			*out_of_puff = true;
275 			break;
276 		}
277 
278 		balloon_page_push(&pages, page);
279 		page_relinquish(page, 1);
280 	}
281 
282 	mutex_lock(&vb->balloon_lock);
283 
284 	vb->num_pfns = 0;
285 
286 	while ((page = balloon_page_pop(&pages))) {
287 		balloon_page_enqueue(&vb->vb_dev_info, page);
288 
289 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
290 		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
291 		if (!virtio_has_feature(vb->vdev,
292 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
293 			adjust_managed_page_count(page, -1);
294 		vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
295 	}
296 
297 	num_allocated_pages = vb->num_pfns;
298 	/* Did we get any? */
299 	if (vb->num_pfns != 0)
300 		tell_host(vb, vb->inflate_vq);
301 	mutex_unlock(&vb->balloon_lock);
302 
303 	return num_allocated_pages;
304 }
305 
release_pages_balloon(struct virtio_balloon * vb,struct list_head * pages)306 static void release_pages_balloon(struct virtio_balloon *vb,
307 				 struct list_head *pages)
308 {
309 	struct page *page, *next;
310 
311 	list_for_each_entry_safe(page, next, pages, lru) {
312 		if (!virtio_has_feature(vb->vdev,
313 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
314 			adjust_managed_page_count(page, 1);
315 		list_del(&page->lru);
316 		put_page(page); /* balloon reference */
317 	}
318 }
319 
leak_balloon(struct virtio_balloon * vb,size_t num)320 static unsigned int leak_balloon(struct virtio_balloon *vb, size_t num)
321 {
322 	unsigned int num_freed_pages;
323 	struct page *page;
324 	struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
325 	LIST_HEAD(pages);
326 
327 	/* We can only do one array worth at a time. */
328 	num = min(num, ARRAY_SIZE(vb->pfns));
329 
330 	mutex_lock(&vb->balloon_lock);
331 	/* We can't release more pages than taken */
332 	num = min(num, (size_t)vb->num_pages);
333 	for (vb->num_pfns = 0; vb->num_pfns < num;
334 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
335 		page = balloon_page_dequeue(vb_dev_info);
336 		if (!page)
337 			break;
338 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
339 		list_add(&page->lru, &pages);
340 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
341 	}
342 
343 	num_freed_pages = vb->num_pfns;
344 	/*
345 	 * Note that if
346 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
347 	 * is true, we *have* to do it in this order
348 	 */
349 	if (vb->num_pfns != 0)
350 		tell_host(vb, vb->deflate_vq);
351 	release_pages_balloon(vb, &pages);
352 	mutex_unlock(&vb->balloon_lock);
353 	return num_freed_pages;
354 }
355 
update_stat(struct virtio_balloon * vb,int idx,u16 tag,u64 val)356 static inline void update_stat(struct virtio_balloon *vb, int idx,
357 			       u16 tag, u64 val)
358 {
359 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
360 	vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
361 	vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
362 }
363 
364 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
365 
366 #ifdef CONFIG_VM_EVENT_COUNTERS
367 /* Return the number of entries filled by vm events */
update_balloon_vm_stats(struct virtio_balloon * vb)368 static inline unsigned int update_balloon_vm_stats(struct virtio_balloon *vb)
369 {
370 	unsigned long events[NR_VM_EVENT_ITEMS];
371 	unsigned int idx = 0;
372 	unsigned int zid;
373 	unsigned long stall = 0;
374 
375 	all_vm_events(events);
376 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
377 		    pages_to_bytes(events[PSWPIN]));
378 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
379 		    pages_to_bytes(events[PSWPOUT]));
380 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
381 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
382 	update_stat(vb, idx++, VIRTIO_BALLOON_S_OOM_KILL, events[OOM_KILL]);
383 
384 	/* sum all the stall events */
385 	for (zid = 0; zid < MAX_NR_ZONES; zid++)
386 		stall += events[ALLOCSTALL_NORMAL - ZONE_NORMAL + zid];
387 
388 	update_stat(vb, idx++, VIRTIO_BALLOON_S_ALLOC_STALL, stall);
389 
390 	update_stat(vb, idx++, VIRTIO_BALLOON_S_ASYNC_SCAN,
391 		    pages_to_bytes(events[PGSCAN_KSWAPD]));
392 	update_stat(vb, idx++, VIRTIO_BALLOON_S_DIRECT_SCAN,
393 		    pages_to_bytes(events[PGSCAN_DIRECT]));
394 	update_stat(vb, idx++, VIRTIO_BALLOON_S_ASYNC_RECLAIM,
395 		    pages_to_bytes(events[PGSTEAL_KSWAPD]));
396 	update_stat(vb, idx++, VIRTIO_BALLOON_S_DIRECT_RECLAIM,
397 		    pages_to_bytes(events[PGSTEAL_DIRECT]));
398 
399 #ifdef CONFIG_HUGETLB_PAGE
400 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC,
401 		    events[HTLB_BUDDY_PGALLOC]);
402 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL,
403 		    events[HTLB_BUDDY_PGALLOC_FAIL]);
404 #endif /* CONFIG_HUGETLB_PAGE */
405 
406 	return idx;
407 }
408 #else /* CONFIG_VM_EVENT_COUNTERS */
update_balloon_vm_stats(struct virtio_balloon * vb)409 static inline unsigned int update_balloon_vm_stats(struct virtio_balloon *vb)
410 {
411 	return 0;
412 }
413 #endif /* CONFIG_VM_EVENT_COUNTERS */
414 
update_balloon_stats(struct virtio_balloon * vb)415 static unsigned int update_balloon_stats(struct virtio_balloon *vb)
416 {
417 	struct sysinfo i;
418 	unsigned int idx;
419 	long available;
420 	unsigned long caches;
421 
422 	idx = update_balloon_vm_stats(vb);
423 
424 	si_meminfo(&i);
425 	available = si_mem_available();
426 	caches = global_node_page_state(NR_FILE_PAGES);
427 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
428 				pages_to_bytes(i.freeram));
429 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
430 				pages_to_bytes(i.totalram));
431 	update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
432 				pages_to_bytes(available));
433 	update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES,
434 				pages_to_bytes(caches));
435 
436 	return idx;
437 }
438 
439 /*
440  * While most virtqueues communicate guest-initiated requests to the hypervisor,
441  * the stats queue operates in reverse.  The driver initializes the virtqueue
442  * with a single buffer.  From that point forward, all conversations consist of
443  * a hypervisor request (a call to this function) which directs us to refill
444  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
445  * we delegate the job to a freezable workqueue that will do the actual work via
446  * stats_handle_request().
447  */
stats_request(struct virtqueue * vq)448 static void stats_request(struct virtqueue *vq)
449 {
450 	struct virtio_balloon *vb = vq->vdev->priv;
451 
452 	spin_lock(&vb->stop_update_lock);
453 	if (!vb->stop_update) {
454 		start_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS);
455 		queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
456 	}
457 	spin_unlock(&vb->stop_update_lock);
458 }
459 
stats_handle_request(struct virtio_balloon * vb)460 static void stats_handle_request(struct virtio_balloon *vb)
461 {
462 	struct virtqueue *vq;
463 	struct scatterlist sg;
464 	unsigned int len, num_stats;
465 
466 	num_stats = update_balloon_stats(vb);
467 
468 	vq = vb->stats_vq;
469 	if (!virtqueue_get_buf(vq, &len))
470 		return;
471 	sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
472 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
473 	virtqueue_kick(vq);
474 }
475 
towards_target(struct virtio_balloon * vb)476 static inline s64 towards_target(struct virtio_balloon *vb)
477 {
478 	s64 target;
479 	u32 num_pages;
480 
481 	/* Legacy balloon config space is LE, unlike all other devices. */
482 	virtio_cread_le(vb->vdev, struct virtio_balloon_config, num_pages,
483 			&num_pages);
484 
485 	/*
486 	 * Aligned up to guest page size to avoid inflating and deflating
487 	 * balloon endlessly.
488 	 */
489 	target = ALIGN(num_pages, VIRTIO_BALLOON_PAGES_PER_PAGE);
490 	return target - vb->num_pages;
491 }
492 
493 /* 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)494 static unsigned long return_free_pages_to_mm(struct virtio_balloon *vb,
495 					     unsigned long num_to_return)
496 {
497 	struct page *page;
498 	unsigned long num_returned;
499 
500 	spin_lock_irq(&vb->free_page_list_lock);
501 	for (num_returned = 0; num_returned < num_to_return; num_returned++) {
502 		page = balloon_page_pop(&vb->free_page_list);
503 		if (!page)
504 			break;
505 		free_pages((unsigned long)page_address(page),
506 			   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
507 	}
508 	vb->num_free_page_blocks -= num_returned;
509 	spin_unlock_irq(&vb->free_page_list_lock);
510 
511 	return num_returned;
512 }
513 
virtio_balloon_queue_free_page_work(struct virtio_balloon * vb)514 static void virtio_balloon_queue_free_page_work(struct virtio_balloon *vb)
515 {
516 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
517 		return;
518 
519 	/* No need to queue the work if the bit was already set. */
520 	if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
521 			     &vb->config_read_bitmap))
522 		return;
523 
524 	queue_work(vb->balloon_wq, &vb->report_free_page_work);
525 }
526 
start_update_balloon_size(struct virtio_balloon * vb)527 static void start_update_balloon_size(struct virtio_balloon *vb)
528 {
529 	start_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST);
530 	queue_work(system_freezable_wq, &vb->update_balloon_size_work);
531 }
532 
virtballoon_changed(struct virtio_device * vdev)533 static void virtballoon_changed(struct virtio_device *vdev)
534 {
535 	struct virtio_balloon *vb = vdev->priv;
536 	unsigned long flags;
537 
538 	spin_lock_irqsave(&vb->stop_update_lock, flags);
539 	if (!vb->stop_update) {
540 		start_update_balloon_size(vb);
541 		virtio_balloon_queue_free_page_work(vb);
542 	}
543 	spin_unlock_irqrestore(&vb->stop_update_lock, flags);
544 }
545 
update_balloon_size(struct virtio_balloon * vb)546 static void update_balloon_size(struct virtio_balloon *vb)
547 {
548 	u32 actual = vb->num_pages;
549 
550 	/* Legacy balloon config space is LE, unlike all other devices. */
551 	virtio_cwrite_le(vb->vdev, struct virtio_balloon_config, actual,
552 			 &actual);
553 }
554 
update_balloon_stats_func(struct work_struct * work)555 static void update_balloon_stats_func(struct work_struct *work)
556 {
557 	struct virtio_balloon *vb;
558 
559 	vb = container_of(work, struct virtio_balloon,
560 			  update_balloon_stats_work);
561 
562 	process_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS);
563 	stats_handle_request(vb);
564 	finish_wakeup_event(vb);
565 }
566 
update_balloon_size_func(struct work_struct * work)567 static void update_balloon_size_func(struct work_struct *work)
568 {
569 	struct virtio_balloon *vb;
570 	s64 diff;
571 	bool out_of_puff = false;
572 
573 	vb = container_of(work, struct virtio_balloon,
574 			  update_balloon_size_work);
575 
576 	process_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST);
577 
578 	diff = towards_target(vb);
579 
580 	if (diff) {
581 		if (diff > 0)
582 			diff -= fill_balloon(vb, diff, &out_of_puff);
583 		else
584 			diff += leak_balloon(vb, -diff);
585 		update_balloon_size(vb);
586 	}
587 
588 	if (diff && !(vb->bail_on_out_of_puff && out_of_puff))
589 		queue_work(system_freezable_wq, work);
590 	else
591 		finish_wakeup_event(vb);
592 }
593 
init_vqs(struct virtio_balloon * vb)594 static int init_vqs(struct virtio_balloon *vb)
595 {
596 	struct virtqueue_info vqs_info[VIRTIO_BALLOON_VQ_MAX] = {};
597 	struct virtqueue *vqs[VIRTIO_BALLOON_VQ_MAX];
598 	int err;
599 
600 	/*
601 	 * Inflateq and deflateq are used unconditionally. The names[]
602 	 * will be NULL if the related feature is not enabled, which will
603 	 * cause no allocation for the corresponding virtqueue in find_vqs.
604 	 */
605 	vqs_info[VIRTIO_BALLOON_VQ_INFLATE].callback = balloon_ack;
606 	vqs_info[VIRTIO_BALLOON_VQ_INFLATE].name = "inflate";
607 	vqs_info[VIRTIO_BALLOON_VQ_DEFLATE].callback = balloon_ack;
608 	vqs_info[VIRTIO_BALLOON_VQ_DEFLATE].name = "deflate";
609 
610 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
611 		vqs_info[VIRTIO_BALLOON_VQ_STATS].name = "stats";
612 		vqs_info[VIRTIO_BALLOON_VQ_STATS].callback = stats_request;
613 	}
614 
615 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
616 		vqs_info[VIRTIO_BALLOON_VQ_FREE_PAGE].name = "free_page_vq";
617 
618 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) {
619 		vqs_info[VIRTIO_BALLOON_VQ_REPORTING].name = "reporting_vq";
620 		vqs_info[VIRTIO_BALLOON_VQ_REPORTING].callback = balloon_ack;
621 	}
622 
623 	err = virtio_find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX, vqs,
624 			      vqs_info, NULL);
625 	if (err)
626 		return err;
627 
628 	vb->inflate_vq = vqs[VIRTIO_BALLOON_VQ_INFLATE];
629 	vb->deflate_vq = vqs[VIRTIO_BALLOON_VQ_DEFLATE];
630 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
631 		struct scatterlist sg;
632 		unsigned int num_stats;
633 		vb->stats_vq = vqs[VIRTIO_BALLOON_VQ_STATS];
634 
635 		/*
636 		 * Prime this virtqueue with one buffer so the hypervisor can
637 		 * use it to signal us later (it can't be broken yet!).
638 		 */
639 		num_stats = update_balloon_stats(vb);
640 
641 		sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
642 		err = virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb,
643 					   GFP_KERNEL);
644 		if (err) {
645 			dev_warn(&vb->vdev->dev, "%s: add stat_vq failed\n",
646 				 __func__);
647 			return err;
648 		}
649 		virtqueue_kick(vb->stats_vq);
650 	}
651 
652 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
653 		vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE];
654 		virtqueue_disable_dma_api_for_buffers(vb->free_page_vq);
655 	}
656 
657 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) {
658 		vb->reporting_vq = vqs[VIRTIO_BALLOON_VQ_REPORTING];
659 		virtqueue_disable_dma_api_for_buffers(vb->reporting_vq);
660 	}
661 
662 	return 0;
663 }
664 
virtio_balloon_cmd_id_received(struct virtio_balloon * vb)665 static u32 virtio_balloon_cmd_id_received(struct virtio_balloon *vb)
666 {
667 	if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
668 			       &vb->config_read_bitmap)) {
669 		/* Legacy balloon config space is LE, unlike all other devices. */
670 		virtio_cread_le(vb->vdev, struct virtio_balloon_config,
671 				free_page_hint_cmd_id,
672 				&vb->cmd_id_received_cache);
673 	}
674 
675 	return vb->cmd_id_received_cache;
676 }
677 
send_cmd_id_start(struct virtio_balloon * vb)678 static int send_cmd_id_start(struct virtio_balloon *vb)
679 {
680 	struct scatterlist sg;
681 	struct virtqueue *vq = vb->free_page_vq;
682 	int err, unused;
683 
684 	/* Detach all the used buffers from the vq */
685 	while (virtqueue_get_buf(vq, &unused))
686 		;
687 
688 	vb->cmd_id_active = cpu_to_virtio32(vb->vdev,
689 					virtio_balloon_cmd_id_received(vb));
690 	sg_init_one(&sg, &vb->cmd_id_active, sizeof(vb->cmd_id_active));
691 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_active, GFP_KERNEL);
692 	if (!err)
693 		virtqueue_kick(vq);
694 	return err;
695 }
696 
send_cmd_id_stop(struct virtio_balloon * vb)697 static int send_cmd_id_stop(struct virtio_balloon *vb)
698 {
699 	struct scatterlist sg;
700 	struct virtqueue *vq = vb->free_page_vq;
701 	int err, unused;
702 
703 	/* Detach all the used buffers from the vq */
704 	while (virtqueue_get_buf(vq, &unused))
705 		;
706 
707 	sg_init_one(&sg, &vb->cmd_id_stop, sizeof(vb->cmd_id_stop));
708 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_stop, GFP_KERNEL);
709 	if (!err)
710 		virtqueue_kick(vq);
711 	return err;
712 }
713 
get_free_page_and_send(struct virtio_balloon * vb)714 static int get_free_page_and_send(struct virtio_balloon *vb)
715 {
716 	struct virtqueue *vq = vb->free_page_vq;
717 	struct page *page;
718 	struct scatterlist sg;
719 	int err, unused;
720 	void *p;
721 
722 	/* Detach all the used buffers from the vq */
723 	while (virtqueue_get_buf(vq, &unused))
724 		;
725 
726 	page = alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG,
727 			   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
728 	/*
729 	 * When the allocation returns NULL, it indicates that we have got all
730 	 * the possible free pages, so return -EINTR to stop.
731 	 */
732 	if (!page)
733 		return -EINTR;
734 
735 	p = page_address(page);
736 	sg_init_one(&sg, p, VIRTIO_BALLOON_HINT_BLOCK_BYTES);
737 	/* There is always 1 entry reserved for the cmd id to use. */
738 	if (vq->num_free > 1) {
739 		err = virtqueue_add_inbuf(vq, &sg, 1, p, GFP_KERNEL);
740 		if (unlikely(err)) {
741 			free_pages((unsigned long)p,
742 				   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
743 			return err;
744 		}
745 		virtqueue_kick(vq);
746 		spin_lock_irq(&vb->free_page_list_lock);
747 		balloon_page_push(&vb->free_page_list, page);
748 		vb->num_free_page_blocks++;
749 		spin_unlock_irq(&vb->free_page_list_lock);
750 	} else {
751 		/*
752 		 * The vq has no available entry to add this page block, so
753 		 * just free it.
754 		 */
755 		free_pages((unsigned long)p, VIRTIO_BALLOON_HINT_BLOCK_ORDER);
756 	}
757 
758 	return 0;
759 }
760 
send_free_pages(struct virtio_balloon * vb)761 static int send_free_pages(struct virtio_balloon *vb)
762 {
763 	int err;
764 	u32 cmd_id_active;
765 
766 	while (1) {
767 		/*
768 		 * If a stop id or a new cmd id was just received from host,
769 		 * stop the reporting.
770 		 */
771 		cmd_id_active = virtio32_to_cpu(vb->vdev, vb->cmd_id_active);
772 		if (unlikely(cmd_id_active !=
773 			     virtio_balloon_cmd_id_received(vb)))
774 			break;
775 
776 		/*
777 		 * The free page blocks are allocated and sent to host one by
778 		 * one.
779 		 */
780 		err = get_free_page_and_send(vb);
781 		if (err == -EINTR)
782 			break;
783 		else if (unlikely(err))
784 			return err;
785 	}
786 
787 	return 0;
788 }
789 
virtio_balloon_report_free_page(struct virtio_balloon * vb)790 static void virtio_balloon_report_free_page(struct virtio_balloon *vb)
791 {
792 	int err;
793 	struct device *dev = &vb->vdev->dev;
794 
795 	/* Start by sending the received cmd id to host with an outbuf. */
796 	err = send_cmd_id_start(vb);
797 	if (unlikely(err))
798 		dev_err(dev, "Failed to send a start id, err = %d\n", err);
799 
800 	err = send_free_pages(vb);
801 	if (unlikely(err))
802 		dev_err(dev, "Failed to send a free page, err = %d\n", err);
803 
804 	/* End by sending a stop id to host with an outbuf. */
805 	err = send_cmd_id_stop(vb);
806 	if (unlikely(err))
807 		dev_err(dev, "Failed to send a stop id, err = %d\n", err);
808 }
809 
report_free_page_func(struct work_struct * work)810 static void report_free_page_func(struct work_struct *work)
811 {
812 	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
813 						 report_free_page_work);
814 	u32 cmd_id_received;
815 
816 	cmd_id_received = virtio_balloon_cmd_id_received(vb);
817 	if (cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
818 		/* Pass ULONG_MAX to give back all the free pages */
819 		return_free_pages_to_mm(vb, ULONG_MAX);
820 	} else if (cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
821 		   cmd_id_received !=
822 		   virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) {
823 		virtio_balloon_report_free_page(vb);
824 	}
825 }
826 
827 #ifdef CONFIG_BALLOON_COMPACTION
828 /*
829  * virtballoon_migratepage - perform the balloon page migration on behalf of
830  *			     a compaction thread.     (called under page lock)
831  * @vb_dev_info: the balloon device
832  * @newpage: page that will replace the isolated page after migration finishes.
833  * @page   : the isolated (old) page that is about to be migrated to newpage.
834  * @mode   : compaction mode -- not used for balloon page migration.
835  *
836  * After a ballooned page gets isolated by compaction procedures, this is the
837  * function that performs the page migration on behalf of a compaction thread
838  * The page migration for virtio balloon is done in a simple swap fashion which
839  * follows these two macro steps:
840  *  1) insert newpage into vb->pages list and update the host about it;
841  *  2) update the host about the old page removed from vb->pages list;
842  *
843  * This function preforms the balloon page migration task.
844  * Called through movable_operations->migrate_page
845  */
virtballoon_migratepage(struct balloon_dev_info * vb_dev_info,struct page * newpage,struct page * page,enum migrate_mode mode)846 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
847 		struct page *newpage, struct page *page, enum migrate_mode mode)
848 {
849 	struct virtio_balloon *vb = container_of(vb_dev_info,
850 			struct virtio_balloon, vb_dev_info);
851 	unsigned long flags;
852 
853 	/*
854 	 * In order to avoid lock contention while migrating pages concurrently
855 	 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
856 	 * this turn, as it is easier to retry the page migration later.
857 	 * This also prevents fill_balloon() getting stuck into a mutex
858 	 * recursion in the case it ends up triggering memory compaction
859 	 * while it is attempting to inflate the ballon.
860 	 */
861 	if (!mutex_trylock(&vb->balloon_lock))
862 		return -EAGAIN;
863 
864 	get_page(newpage); /* balloon reference */
865 
866 	/*
867 	  * When we migrate a page to a different zone and adjusted the
868 	  * managed page count when inflating, we have to fixup the count of
869 	  * both involved zones.
870 	  */
871 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM) &&
872 	    page_zone(page) != page_zone(newpage)) {
873 		adjust_managed_page_count(page, 1);
874 		adjust_managed_page_count(newpage, -1);
875 	}
876 
877 	/* balloon's page migration 1st step  -- inflate "newpage" */
878 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
879 	balloon_page_insert(vb_dev_info, newpage);
880 	vb_dev_info->isolated_pages--;
881 	__count_vm_event(BALLOON_MIGRATE);
882 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
883 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
884 	set_page_pfns(vb, vb->pfns, newpage);
885 	tell_host(vb, vb->inflate_vq);
886 
887 	/* balloon's page migration 2nd step -- deflate "page" */
888 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
889 	balloon_page_delete(page);
890 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
891 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
892 	set_page_pfns(vb, vb->pfns, page);
893 	tell_host(vb, vb->deflate_vq);
894 
895 	mutex_unlock(&vb->balloon_lock);
896 
897 	put_page(page); /* balloon reference */
898 
899 	return MIGRATEPAGE_SUCCESS;
900 }
901 #endif /* CONFIG_BALLOON_COMPACTION */
902 
shrink_free_pages(struct virtio_balloon * vb,unsigned long pages_to_free)903 static unsigned long shrink_free_pages(struct virtio_balloon *vb,
904 				       unsigned long pages_to_free)
905 {
906 	unsigned long blocks_to_free, blocks_freed;
907 
908 	pages_to_free = round_up(pages_to_free,
909 				 VIRTIO_BALLOON_HINT_BLOCK_PAGES);
910 	blocks_to_free = pages_to_free / VIRTIO_BALLOON_HINT_BLOCK_PAGES;
911 	blocks_freed = return_free_pages_to_mm(vb, blocks_to_free);
912 
913 	return blocks_freed * VIRTIO_BALLOON_HINT_BLOCK_PAGES;
914 }
915 
virtio_balloon_shrinker_scan(struct shrinker * shrinker,struct shrink_control * sc)916 static unsigned long virtio_balloon_shrinker_scan(struct shrinker *shrinker,
917 						  struct shrink_control *sc)
918 {
919 	struct virtio_balloon *vb = shrinker->private_data;
920 
921 	return shrink_free_pages(vb, sc->nr_to_scan);
922 }
923 
virtio_balloon_shrinker_count(struct shrinker * shrinker,struct shrink_control * sc)924 static unsigned long virtio_balloon_shrinker_count(struct shrinker *shrinker,
925 						   struct shrink_control *sc)
926 {
927 	struct virtio_balloon *vb = shrinker->private_data;
928 
929 	return vb->num_free_page_blocks * VIRTIO_BALLOON_HINT_BLOCK_PAGES;
930 }
931 
virtio_balloon_oom_notify(struct notifier_block * nb,unsigned long dummy,void * parm)932 static int virtio_balloon_oom_notify(struct notifier_block *nb,
933 				     unsigned long dummy, void *parm)
934 {
935 	struct virtio_balloon *vb = container_of(nb,
936 						 struct virtio_balloon, oom_nb);
937 	unsigned long *freed = parm;
938 
939 	*freed += leak_balloon(vb, VIRTIO_BALLOON_OOM_NR_PAGES) /
940 		  VIRTIO_BALLOON_PAGES_PER_PAGE;
941 	update_balloon_size(vb);
942 
943 	return NOTIFY_OK;
944 }
945 
virtio_balloon_unregister_shrinker(struct virtio_balloon * vb)946 static void virtio_balloon_unregister_shrinker(struct virtio_balloon *vb)
947 {
948 	shrinker_free(vb->shrinker);
949 }
950 
virtio_balloon_register_shrinker(struct virtio_balloon * vb)951 static int virtio_balloon_register_shrinker(struct virtio_balloon *vb)
952 {
953 	vb->shrinker = shrinker_alloc(0, "virtio-balloon");
954 	if (!vb->shrinker)
955 		return -ENOMEM;
956 
957 	vb->shrinker->scan_objects = virtio_balloon_shrinker_scan;
958 	vb->shrinker->count_objects = virtio_balloon_shrinker_count;
959 	vb->shrinker->private_data = vb;
960 
961 	shrinker_register(vb->shrinker);
962 
963 	return 0;
964 }
965 
bail_on_out_of_puff_show(struct device * d,struct device_attribute * attr,char * buf)966 static ssize_t bail_on_out_of_puff_show(struct device *d, struct device_attribute *attr,
967 			       char *buf)
968 {
969 	struct virtio_device *vdev =
970 		container_of(d, struct virtio_device, dev);
971 	struct virtio_balloon *vb = vdev->priv;
972 
973 	return sprintf(buf, "%c\n", vb->bail_on_out_of_puff ? '1' : '0');
974 }
975 
bail_on_out_of_puff_store(struct device * d,struct device_attribute * attr,const char * buf,size_t count)976 static ssize_t bail_on_out_of_puff_store(struct device *d, struct device_attribute *attr,
977 				const char *buf, size_t count)
978 {
979 	struct virtio_device *vdev =
980 		container_of(d, struct virtio_device, dev);
981 	struct virtio_balloon *vb = vdev->priv;
982 
983 	return kstrtobool(buf, &vb->bail_on_out_of_puff) ?: count;
984 }
985 
986 static DEVICE_ATTR_RW(bail_on_out_of_puff);
987 
988 static struct attribute *virtio_balloon_sysfs_entries[] = {
989 	&dev_attr_bail_on_out_of_puff.attr,
990 	NULL
991 };
992 
993 static const struct attribute_group virtio_balloon_attribute_group = {
994 	.name = NULL,		/* put in device directory */
995 	.attrs = virtio_balloon_sysfs_entries,
996 };
997 
virtballoon_probe(struct virtio_device * vdev)998 static int virtballoon_probe(struct virtio_device *vdev)
999 {
1000 	struct virtio_balloon *vb;
1001 	int err;
1002 
1003 	if (!vdev->config->get) {
1004 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
1005 			__func__);
1006 		return -EINVAL;
1007 	}
1008 
1009 	vdev->priv = vb = kzalloc(sizeof(*vb), GFP_KERNEL);
1010 	if (!vb) {
1011 		err = -ENOMEM;
1012 		goto out;
1013 	}
1014 
1015 	INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
1016 	INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
1017 	spin_lock_init(&vb->stop_update_lock);
1018 	mutex_init(&vb->balloon_lock);
1019 	init_waitqueue_head(&vb->acked);
1020 	vb->vdev = vdev;
1021 
1022 	balloon_devinfo_init(&vb->vb_dev_info);
1023 
1024 	err = init_vqs(vb);
1025 	if (err)
1026 		goto out_free_vb;
1027 
1028 	err = sysfs_create_group(&vdev->dev.kobj,
1029 				 &virtio_balloon_attribute_group);
1030 	if (err)
1031 		goto out_del_vqs;
1032 
1033 #ifdef CONFIG_BALLOON_COMPACTION
1034 	vb->vb_dev_info.migratepage = virtballoon_migratepage;
1035 #endif
1036 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
1037 		/*
1038 		 * There is always one entry reserved for cmd id, so the ring
1039 		 * size needs to be at least two to report free page hints.
1040 		 */
1041 		if (virtqueue_get_vring_size(vb->free_page_vq) < 2) {
1042 			err = -ENOSPC;
1043 			goto out_remove_sysfs;
1044 		}
1045 		vb->balloon_wq = alloc_workqueue("balloon-wq",
1046 					WQ_FREEZABLE | WQ_CPU_INTENSIVE, 0);
1047 		if (!vb->balloon_wq) {
1048 			err = -ENOMEM;
1049 			goto out_remove_sysfs;
1050 		}
1051 		INIT_WORK(&vb->report_free_page_work, report_free_page_func);
1052 		vb->cmd_id_received_cache = VIRTIO_BALLOON_CMD_ID_STOP;
1053 		vb->cmd_id_active = cpu_to_virtio32(vb->vdev,
1054 						  VIRTIO_BALLOON_CMD_ID_STOP);
1055 		vb->cmd_id_stop = cpu_to_virtio32(vb->vdev,
1056 						  VIRTIO_BALLOON_CMD_ID_STOP);
1057 		spin_lock_init(&vb->free_page_list_lock);
1058 		INIT_LIST_HEAD(&vb->free_page_list);
1059 		/*
1060 		 * We're allowed to reuse any free pages, even if they are
1061 		 * still to be processed by the host.
1062 		 */
1063 		err = virtio_balloon_register_shrinker(vb);
1064 		if (err)
1065 			goto out_del_balloon_wq;
1066 	}
1067 
1068 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) {
1069 		vb->oom_nb.notifier_call = virtio_balloon_oom_notify;
1070 		vb->oom_nb.priority = VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY;
1071 		err = register_oom_notifier(&vb->oom_nb);
1072 		if (err < 0)
1073 			goto out_unregister_shrinker;
1074 	}
1075 
1076 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) {
1077 		/* Start with poison val of 0 representing general init */
1078 		__u32 poison_val = 0;
1079 
1080 		/*
1081 		 * Let the hypervisor know that we are expecting a
1082 		 * specific value to be written back in balloon pages.
1083 		 *
1084 		 * If the PAGE_POISON value was larger than a byte we would
1085 		 * need to byte swap poison_val here to guarantee it is
1086 		 * little-endian. However for now it is a single byte so we
1087 		 * can pass it as-is.
1088 		 */
1089 		if (!want_init_on_free())
1090 			memset(&poison_val, PAGE_POISON, sizeof(poison_val));
1091 
1092 		virtio_cwrite_le(vb->vdev, struct virtio_balloon_config,
1093 				 poison_val, &poison_val);
1094 	}
1095 
1096 	vb->pr_dev_info.report = virtballoon_free_page_report;
1097 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) {
1098 		unsigned int capacity;
1099 
1100 		capacity = virtqueue_get_vring_size(vb->reporting_vq);
1101 		if (capacity < PAGE_REPORTING_CAPACITY) {
1102 			err = -ENOSPC;
1103 			goto out_unregister_oom;
1104 		}
1105 
1106 		/*
1107 		 * The default page reporting order is @pageblock_order, which
1108 		 * corresponds to 512MB in size on ARM64 when 64KB base page
1109 		 * size is used. The page reporting won't be triggered if the
1110 		 * freeing page can't come up with a free area like that huge.
1111 		 * So we specify the page reporting order to 5, corresponding
1112 		 * to 2MB. It helps to avoid THP splitting if 4KB base page
1113 		 * size is used by host.
1114 		 *
1115 		 * Ideally, the page reporting order is selected based on the
1116 		 * host's base page size. However, it needs more work to report
1117 		 * that value. The hard-coded order would be fine currently.
1118 		 */
1119 #if defined(CONFIG_ARM64) && defined(CONFIG_ARM64_64K_PAGES)
1120 		vb->pr_dev_info.order = 5;
1121 #endif
1122 
1123 		err = page_reporting_register(&vb->pr_dev_info);
1124 		if (err)
1125 			goto out_unregister_oom;
1126 	}
1127 
1128 	spin_lock_init(&vb->wakeup_lock);
1129 
1130 	/*
1131 	 * The virtio balloon itself can't wake up the device, but it is
1132 	 * responsible for processing wakeup events passed up from the transport
1133 	 * layer. Wakeup sources don't support nesting/chaining calls, so we use
1134 	 * our own wakeup source to ensure wakeup events are properly handled
1135 	 * without trampling on the transport layer's wakeup source.
1136 	 */
1137 	device_set_wakeup_capable(&vb->vdev->dev, true);
1138 
1139 	virtio_device_ready(vdev);
1140 
1141 	if (towards_target(vb))
1142 		virtballoon_changed(vdev);
1143 	return 0;
1144 
1145 out_unregister_oom:
1146 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
1147 		unregister_oom_notifier(&vb->oom_nb);
1148 out_unregister_shrinker:
1149 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1150 		virtio_balloon_unregister_shrinker(vb);
1151 out_del_balloon_wq:
1152 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1153 		destroy_workqueue(vb->balloon_wq);
1154 out_remove_sysfs:
1155 	sysfs_remove_group(&vdev->dev.kobj, &virtio_balloon_attribute_group);
1156 out_del_vqs:
1157 	vdev->config->del_vqs(vdev);
1158 out_free_vb:
1159 	kfree(vb);
1160 out:
1161 	return err;
1162 }
1163 
remove_common(struct virtio_balloon * vb)1164 static void remove_common(struct virtio_balloon *vb)
1165 {
1166 	/* There might be pages left in the balloon: free them. */
1167 	while (vb->num_pages)
1168 		leak_balloon(vb, vb->num_pages);
1169 	update_balloon_size(vb);
1170 
1171 	/* There might be free pages that are being reported: release them. */
1172 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1173 		return_free_pages_to_mm(vb, ULONG_MAX);
1174 
1175 	/* Now we reset the device so we can clean up the queues. */
1176 	virtio_reset_device(vb->vdev);
1177 
1178 	vb->vdev->config->del_vqs(vb->vdev);
1179 }
1180 
virtballoon_remove(struct virtio_device * vdev)1181 static void virtballoon_remove(struct virtio_device *vdev)
1182 {
1183 	struct virtio_balloon *vb = vdev->priv;
1184 
1185 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING))
1186 		page_reporting_unregister(&vb->pr_dev_info);
1187 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
1188 		unregister_oom_notifier(&vb->oom_nb);
1189 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1190 		virtio_balloon_unregister_shrinker(vb);
1191 	spin_lock_irq(&vb->stop_update_lock);
1192 	vb->stop_update = true;
1193 	spin_unlock_irq(&vb->stop_update_lock);
1194 	cancel_work_sync(&vb->update_balloon_size_work);
1195 	cancel_work_sync(&vb->update_balloon_stats_work);
1196 
1197 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
1198 		cancel_work_sync(&vb->report_free_page_work);
1199 		destroy_workqueue(vb->balloon_wq);
1200 	}
1201 
1202 	sysfs_remove_group(&vdev->dev.kobj, &virtio_balloon_attribute_group);
1203 
1204 	remove_common(vb);
1205 	kfree(vb);
1206 }
1207 
1208 #ifdef CONFIG_PM_SLEEP
virtballoon_freeze(struct virtio_device * vdev)1209 static int virtballoon_freeze(struct virtio_device *vdev)
1210 {
1211 	struct virtio_balloon *vb = vdev->priv;
1212 
1213 	/*
1214 	 * The workqueue is already frozen by the PM core before this
1215 	 * function is called.
1216 	 */
1217 	remove_common(vb);
1218 	return 0;
1219 }
1220 
virtballoon_restore(struct virtio_device * vdev)1221 static int virtballoon_restore(struct virtio_device *vdev)
1222 {
1223 	struct virtio_balloon *vb = vdev->priv;
1224 	int ret;
1225 
1226 	ret = init_vqs(vdev->priv);
1227 	if (ret)
1228 		return ret;
1229 
1230 	virtio_device_ready(vdev);
1231 
1232 	if (towards_target(vb))
1233 		virtballoon_changed(vdev);
1234 	update_balloon_size(vb);
1235 	return 0;
1236 }
1237 #endif
1238 
virtballoon_validate(struct virtio_device * vdev)1239 static int virtballoon_validate(struct virtio_device *vdev)
1240 {
1241 	if (WARN_ON(page_relinquish_disallowed()))
1242 		return -EINVAL;
1243 
1244 	/*
1245 	 * Inform the hypervisor that our pages are poisoned or
1246 	 * initialized. If we cannot do that then we should disable
1247 	 * page reporting as it could potentially change the contents
1248 	 * of our free pages.
1249 	 */
1250 	if (!want_init_on_free() && !page_poisoning_enabled_static())
1251 		__virtio_clear_bit(vdev, VIRTIO_BALLOON_F_PAGE_POISON);
1252 	else if (!virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON))
1253 		__virtio_clear_bit(vdev, VIRTIO_BALLOON_F_REPORTING);
1254 
1255 	return 0;
1256 }
1257 
1258 static unsigned int features[] = {
1259 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
1260 	VIRTIO_BALLOON_F_STATS_VQ,
1261 	VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
1262 	VIRTIO_BALLOON_F_FREE_PAGE_HINT,
1263 	VIRTIO_BALLOON_F_PAGE_POISON,
1264 	VIRTIO_BALLOON_F_REPORTING,
1265 };
1266 
1267 static struct virtio_driver virtio_balloon_driver = {
1268 	.feature_table = features,
1269 	.feature_table_size = ARRAY_SIZE(features),
1270 	.driver.name =	KBUILD_MODNAME,
1271 	.id_table =	id_table,
1272 	.validate =	virtballoon_validate,
1273 	.probe =	virtballoon_probe,
1274 	.remove =	virtballoon_remove,
1275 	.config_changed = virtballoon_changed,
1276 #ifdef CONFIG_PM_SLEEP
1277 	.freeze	=	virtballoon_freeze,
1278 	.restore =	virtballoon_restore,
1279 #endif
1280 };
1281 
1282 module_virtio_driver(virtio_balloon_driver);
1283 MODULE_DEVICE_TABLE(virtio, id_table);
1284 MODULE_DESCRIPTION("Virtio balloon driver");
1285 MODULE_LICENSE("GPL");
1286