1/*
2 * Virtio balloon implementation, inspired by Dor Laor and Marcelo
3 * Tosatti's implementations.
4 *
5 *  Copyright 2008 Rusty Russell IBM Corporation
6 *
7 *  This program is free software; you can redistribute it and/or modify
8 *  it under the terms of the GNU General Public License as published by
9 *  the Free Software Foundation; either version 2 of the License, or
10 *  (at your option) any later version.
11 *
12 *  This program is distributed in the hope that it will be useful,
13 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
14 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15 *  GNU General Public License for more details.
16 *
17 *  You should have received a copy of the GNU General Public License
18 *  along with this program; if not, write to the Free Software
19 *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
20 */
21
22#include <linux/virtio.h>
23#include <linux/virtio_balloon.h>
24#include <linux/swap.h>
25#include <linux/kthread.h>
26#include <linux/freezer.h>
27#include <linux/delay.h>
28#include <linux/slab.h>
29#include <linux/module.h>
30#include <linux/balloon_compaction.h>
31#include <linux/oom.h>
32#include <linux/wait.h>
33
34/*
35 * Balloon device works in 4K page units.  So each page is pointed to by
36 * multiple balloon pages.  All memory counters in this driver are in balloon
37 * page units.
38 */
39#define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
40#define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
41#define OOM_VBALLOON_DEFAULT_PAGES 256
42#define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
43
44static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES;
45module_param(oom_pages, int, S_IRUSR | S_IWUSR);
46MODULE_PARM_DESC(oom_pages, "pages to free on OOM");
47
48struct virtio_balloon {
49	struct virtio_device *vdev;
50	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
51
52	/* Where the ballooning thread waits for config to change. */
53	wait_queue_head_t config_change;
54
55	/* The thread servicing the balloon. */
56	struct task_struct *thread;
57
58	/* Waiting for host to ack the pages we released. */
59	wait_queue_head_t acked;
60
61	/* Number of balloon pages we've told the Host we're not using. */
62	unsigned int num_pages;
63	/*
64	 * The pages we've told the Host we're not using are enqueued
65	 * at vb_dev_info->pages list.
66	 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
67	 * to num_pages above.
68	 */
69	struct balloon_dev_info vb_dev_info;
70
71	/* Synchronize access/update to this struct virtio_balloon elements */
72	struct mutex balloon_lock;
73
74	/* The array of pfns we tell the Host about. */
75	unsigned int num_pfns;
76	u32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
77
78	/* Memory statistics */
79	int need_stats_update;
80	struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
81
82	/* To register callback in oom notifier call chain */
83	struct notifier_block nb;
84};
85
86static struct virtio_device_id id_table[] = {
87	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
88	{ 0 },
89};
90
91static u32 page_to_balloon_pfn(struct page *page)
92{
93	unsigned long pfn = page_to_pfn(page);
94
95	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
96	/* Convert pfn from Linux page size to balloon page size. */
97	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
98}
99
100static struct page *balloon_pfn_to_page(u32 pfn)
101{
102	BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE);
103	return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE);
104}
105
106static void balloon_ack(struct virtqueue *vq)
107{
108	struct virtio_balloon *vb = vq->vdev->priv;
109
110	wake_up(&vb->acked);
111}
112
113static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
114{
115	struct scatterlist sg;
116	unsigned int len;
117
118	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
119
120	/* We should always be able to add one buffer to an empty queue. */
121	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
122	virtqueue_kick(vq);
123
124	/* When host has read buffer, this completes via balloon_ack */
125	wait_event(vb->acked, virtqueue_get_buf(vq, &len));
126}
127
128static void set_page_pfns(u32 pfns[], struct page *page)
129{
130	unsigned int i;
131
132	/* Set balloon pfns pointing at this page.
133	 * Note that the first pfn points at start of the page. */
134	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
135		pfns[i] = page_to_balloon_pfn(page) + i;
136}
137
138static void fill_balloon(struct virtio_balloon *vb, size_t num)
139{
140	struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
141
142	/* We can only do one array worth at a time. */
143	num = min(num, ARRAY_SIZE(vb->pfns));
144
145	mutex_lock(&vb->balloon_lock);
146	for (vb->num_pfns = 0; vb->num_pfns < num;
147	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
148		struct page *page = balloon_page_enqueue(vb_dev_info);
149
150		if (!page) {
151			dev_info_ratelimited(&vb->vdev->dev,
152					     "Out of puff! Can't get %u pages\n",
153					     VIRTIO_BALLOON_PAGES_PER_PAGE);
154			/* Sleep for at least 1/5 of a second before retry. */
155			msleep(200);
156			break;
157		}
158		set_page_pfns(vb->pfns + vb->num_pfns, page);
159		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
160		adjust_managed_page_count(page, -1);
161	}
162
163	/* Did we get any? */
164	if (vb->num_pfns != 0)
165		tell_host(vb, vb->inflate_vq);
166	mutex_unlock(&vb->balloon_lock);
167}
168
169static void release_pages_balloon(struct virtio_balloon *vb)
170{
171	unsigned int i;
172
173	/* Find pfns pointing at start of each page, get pages and free them. */
174	for (i = 0; i < vb->num_pfns; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
175		struct page *page = balloon_pfn_to_page(vb->pfns[i]);
176		adjust_managed_page_count(page, 1);
177		put_page(page); /* balloon reference */
178	}
179}
180
181static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
182{
183	unsigned num_freed_pages;
184	struct page *page;
185	struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
186
187	/* We can only do one array worth at a time. */
188	num = min(num, ARRAY_SIZE(vb->pfns));
189
190	mutex_lock(&vb->balloon_lock);
191	for (vb->num_pfns = 0; vb->num_pfns < num;
192	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
193		page = balloon_page_dequeue(vb_dev_info);
194		if (!page)
195			break;
196		set_page_pfns(vb->pfns + vb->num_pfns, page);
197		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
198	}
199
200	num_freed_pages = vb->num_pfns;
201	/*
202	 * Note that if
203	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
204	 * is true, we *have* to do it in this order
205	 */
206	if (vb->num_pfns != 0)
207		tell_host(vb, vb->deflate_vq);
208	release_pages_balloon(vb);
209	mutex_unlock(&vb->balloon_lock);
210	return num_freed_pages;
211}
212
213static inline void update_stat(struct virtio_balloon *vb, int idx,
214			       u16 tag, u64 val)
215{
216	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
217	vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
218	vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
219}
220
221#define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
222
223static void update_balloon_stats(struct virtio_balloon *vb)
224{
225	unsigned long events[NR_VM_EVENT_ITEMS];
226	struct sysinfo i;
227	int idx = 0;
228
229	all_vm_events(events);
230	si_meminfo(&i);
231
232	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
233				pages_to_bytes(events[PSWPIN]));
234	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
235				pages_to_bytes(events[PSWPOUT]));
236	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
237	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
238	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
239				pages_to_bytes(i.freeram));
240	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
241				pages_to_bytes(i.totalram));
242}
243
244/*
245 * While most virtqueues communicate guest-initiated requests to the hypervisor,
246 * the stats queue operates in reverse.  The driver initializes the virtqueue
247 * with a single buffer.  From that point forward, all conversations consist of
248 * a hypervisor request (a call to this function) which directs us to refill
249 * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
250 * we notify our kthread which does the actual work via stats_handle_request().
251 */
252static void stats_request(struct virtqueue *vq)
253{
254	struct virtio_balloon *vb = vq->vdev->priv;
255
256	vb->need_stats_update = 1;
257	wake_up(&vb->config_change);
258}
259
260static void stats_handle_request(struct virtio_balloon *vb)
261{
262	struct virtqueue *vq;
263	struct scatterlist sg;
264	unsigned int len;
265
266	vb->need_stats_update = 0;
267	update_balloon_stats(vb);
268
269	vq = vb->stats_vq;
270	if (!virtqueue_get_buf(vq, &len))
271		return;
272	sg_init_one(&sg, vb->stats, sizeof(vb->stats));
273	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
274	virtqueue_kick(vq);
275}
276
277static void virtballoon_changed(struct virtio_device *vdev)
278{
279	struct virtio_balloon *vb = vdev->priv;
280
281	wake_up(&vb->config_change);
282}
283
284static inline s64 towards_target(struct virtio_balloon *vb)
285{
286	s64 target;
287	u32 num_pages;
288
289	virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
290		     &num_pages);
291
292	/* Legacy balloon config space is LE, unlike all other devices. */
293	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
294		num_pages = le32_to_cpu((__force __le32)num_pages);
295
296	target = num_pages;
297	return target - vb->num_pages;
298}
299
300static void update_balloon_size(struct virtio_balloon *vb)
301{
302	u32 actual = vb->num_pages;
303
304	/* Legacy balloon config space is LE, unlike all other devices. */
305	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
306		actual = (__force u32)cpu_to_le32(actual);
307
308	virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
309		      &actual);
310}
311
312/*
313 * virtballoon_oom_notify - release pages when system is under severe
314 *			    memory pressure (called from out_of_memory())
315 * @self : notifier block struct
316 * @dummy: not used
317 * @parm : returned - number of freed pages
318 *
319 * The balancing of memory by use of the virtio balloon should not cause
320 * the termination of processes while there are pages in the balloon.
321 * If virtio balloon manages to release some memory, it will make the
322 * system return and retry the allocation that forced the OOM killer
323 * to run.
324 */
325static int virtballoon_oom_notify(struct notifier_block *self,
326				  unsigned long dummy, void *parm)
327{
328	struct virtio_balloon *vb;
329	unsigned long *freed;
330	unsigned num_freed_pages;
331
332	vb = container_of(self, struct virtio_balloon, nb);
333	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
334		return NOTIFY_OK;
335
336	freed = parm;
337	num_freed_pages = leak_balloon(vb, oom_pages);
338	update_balloon_size(vb);
339	*freed += num_freed_pages;
340
341	return NOTIFY_OK;
342}
343
344static int balloon(void *_vballoon)
345{
346	struct virtio_balloon *vb = _vballoon;
347	DEFINE_WAIT_FUNC(wait, woken_wake_function);
348
349	set_freezable();
350	while (!kthread_should_stop()) {
351		s64 diff;
352
353		try_to_freeze();
354
355		add_wait_queue(&vb->config_change, &wait);
356		for (;;) {
357			if ((diff = towards_target(vb)) != 0 ||
358			    vb->need_stats_update ||
359			    kthread_should_stop() ||
360			    freezing(current))
361				break;
362			wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
363		}
364		remove_wait_queue(&vb->config_change, &wait);
365
366		if (vb->need_stats_update)
367			stats_handle_request(vb);
368		if (diff > 0)
369			fill_balloon(vb, diff);
370		else if (diff < 0)
371			leak_balloon(vb, -diff);
372		update_balloon_size(vb);
373
374		/*
375		 * For large balloon changes, we could spend a lot of time
376		 * and always have work to do.  Be nice if preempt disabled.
377		 */
378		cond_resched();
379	}
380	return 0;
381}
382
383static int init_vqs(struct virtio_balloon *vb)
384{
385	struct virtqueue *vqs[3];
386	vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
387	const char *names[] = { "inflate", "deflate", "stats" };
388	int err, nvqs;
389
390	/*
391	 * We expect two virtqueues: inflate and deflate, and
392	 * optionally stat.
393	 */
394	nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
395	err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
396	if (err)
397		return err;
398
399	vb->inflate_vq = vqs[0];
400	vb->deflate_vq = vqs[1];
401	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
402		struct scatterlist sg;
403		vb->stats_vq = vqs[2];
404
405		/*
406		 * Prime this virtqueue with one buffer so the hypervisor can
407		 * use it to signal us later (it can't be broken yet!).
408		 */
409		sg_init_one(&sg, vb->stats, sizeof vb->stats);
410		if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
411		    < 0)
412			BUG();
413		virtqueue_kick(vb->stats_vq);
414	}
415	return 0;
416}
417
418#ifdef CONFIG_BALLOON_COMPACTION
419/*
420 * virtballoon_migratepage - perform the balloon page migration on behalf of
421 *			     a compation thread.     (called under page lock)
422 * @vb_dev_info: the balloon device
423 * @newpage: page that will replace the isolated page after migration finishes.
424 * @page   : the isolated (old) page that is about to be migrated to newpage.
425 * @mode   : compaction mode -- not used for balloon page migration.
426 *
427 * After a ballooned page gets isolated by compaction procedures, this is the
428 * function that performs the page migration on behalf of a compaction thread
429 * The page migration for virtio balloon is done in a simple swap fashion which
430 * follows these two macro steps:
431 *  1) insert newpage into vb->pages list and update the host about it;
432 *  2) update the host about the old page removed from vb->pages list;
433 *
434 * This function preforms the balloon page migration task.
435 * Called through balloon_mapping->a_ops->migratepage
436 */
437static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
438		struct page *newpage, struct page *page, enum migrate_mode mode)
439{
440	struct virtio_balloon *vb = container_of(vb_dev_info,
441			struct virtio_balloon, vb_dev_info);
442	unsigned long flags;
443
444	/*
445	 * In order to avoid lock contention while migrating pages concurrently
446	 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
447	 * this turn, as it is easier to retry the page migration later.
448	 * This also prevents fill_balloon() getting stuck into a mutex
449	 * recursion in the case it ends up triggering memory compaction
450	 * while it is attempting to inflate the ballon.
451	 */
452	if (!mutex_trylock(&vb->balloon_lock))
453		return -EAGAIN;
454
455	get_page(newpage); /* balloon reference */
456
457	/* balloon's page migration 1st step  -- inflate "newpage" */
458	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
459	balloon_page_insert(vb_dev_info, newpage);
460	vb_dev_info->isolated_pages--;
461	__count_vm_event(BALLOON_MIGRATE);
462	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
463	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
464	set_page_pfns(vb->pfns, newpage);
465	tell_host(vb, vb->inflate_vq);
466
467	/* balloon's page migration 2nd step -- deflate "page" */
468	balloon_page_delete(page);
469	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
470	set_page_pfns(vb->pfns, page);
471	tell_host(vb, vb->deflate_vq);
472
473	mutex_unlock(&vb->balloon_lock);
474
475	put_page(page); /* balloon reference */
476
477	return MIGRATEPAGE_SUCCESS;
478}
479#endif /* CONFIG_BALLOON_COMPACTION */
480
481static int virtballoon_probe(struct virtio_device *vdev)
482{
483	struct virtio_balloon *vb;
484	int err;
485
486	if (!vdev->config->get) {
487		dev_err(&vdev->dev, "%s failure: config access disabled\n",
488			__func__);
489		return -EINVAL;
490	}
491
492	vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
493	if (!vb) {
494		err = -ENOMEM;
495		goto out;
496	}
497
498	vb->num_pages = 0;
499	mutex_init(&vb->balloon_lock);
500	init_waitqueue_head(&vb->config_change);
501	init_waitqueue_head(&vb->acked);
502	vb->vdev = vdev;
503	vb->need_stats_update = 0;
504
505	balloon_devinfo_init(&vb->vb_dev_info);
506#ifdef CONFIG_BALLOON_COMPACTION
507	vb->vb_dev_info.migratepage = virtballoon_migratepage;
508#endif
509
510	err = init_vqs(vb);
511	if (err)
512		goto out_free_vb;
513
514	vb->nb.notifier_call = virtballoon_oom_notify;
515	vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
516	err = register_oom_notifier(&vb->nb);
517	if (err < 0)
518		goto out_oom_notify;
519
520	virtio_device_ready(vdev);
521
522	vb->thread = kthread_run(balloon, vb, "vballoon");
523	if (IS_ERR(vb->thread)) {
524		err = PTR_ERR(vb->thread);
525		goto out_del_vqs;
526	}
527
528	return 0;
529
530out_del_vqs:
531	unregister_oom_notifier(&vb->nb);
532out_oom_notify:
533	vdev->config->del_vqs(vdev);
534out_free_vb:
535	kfree(vb);
536out:
537	return err;
538}
539
540static void remove_common(struct virtio_balloon *vb)
541{
542	/* There might be pages left in the balloon: free them. */
543	while (vb->num_pages)
544		leak_balloon(vb, vb->num_pages);
545	update_balloon_size(vb);
546
547	/* Now we reset the device so we can clean up the queues. */
548	vb->vdev->config->reset(vb->vdev);
549
550	vb->vdev->config->del_vqs(vb->vdev);
551}
552
553static void virtballoon_remove(struct virtio_device *vdev)
554{
555	struct virtio_balloon *vb = vdev->priv;
556
557	unregister_oom_notifier(&vb->nb);
558	kthread_stop(vb->thread);
559	remove_common(vb);
560	kfree(vb);
561}
562
563#ifdef CONFIG_PM_SLEEP
564static int virtballoon_freeze(struct virtio_device *vdev)
565{
566	struct virtio_balloon *vb = vdev->priv;
567
568	/*
569	 * The kthread is already frozen by the PM core before this
570	 * function is called.
571	 */
572
573	remove_common(vb);
574	return 0;
575}
576
577static int virtballoon_restore(struct virtio_device *vdev)
578{
579	struct virtio_balloon *vb = vdev->priv;
580	int ret;
581
582	ret = init_vqs(vdev->priv);
583	if (ret)
584		return ret;
585
586	virtio_device_ready(vdev);
587
588	fill_balloon(vb, towards_target(vb));
589	update_balloon_size(vb);
590	return 0;
591}
592#endif
593
594static unsigned int features[] = {
595	VIRTIO_BALLOON_F_MUST_TELL_HOST,
596	VIRTIO_BALLOON_F_STATS_VQ,
597	VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
598};
599
600static struct virtio_driver virtio_balloon_driver = {
601	.feature_table = features,
602	.feature_table_size = ARRAY_SIZE(features),
603	.driver.name =	KBUILD_MODNAME,
604	.driver.owner =	THIS_MODULE,
605	.id_table =	id_table,
606	.probe =	virtballoon_probe,
607	.remove =	virtballoon_remove,
608	.config_changed = virtballoon_changed,
609#ifdef CONFIG_PM_SLEEP
610	.freeze	=	virtballoon_freeze,
611	.restore =	virtballoon_restore,
612#endif
613};
614
615module_virtio_driver(virtio_balloon_driver);
616MODULE_DEVICE_TABLE(virtio, id_table);
617MODULE_DESCRIPTION("Virtio balloon driver");
618MODULE_LICENSE("GPL");
619