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 
31 /*
32  * Balloon device works in 4K page units.  So each page is pointed to by
33  * multiple balloon pages.  All memory counters in this driver are in balloon
34  * page units.
35  */
36 #define VIRTIO_BALLOON_PAGES_PER_PAGE (PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
37 
38 struct virtio_balloon
39 {
40 	struct virtio_device *vdev;
41 	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
42 
43 	/* Where the ballooning thread waits for config to change. */
44 	wait_queue_head_t config_change;
45 
46 	/* The thread servicing the balloon. */
47 	struct task_struct *thread;
48 
49 	/* Waiting for host to ack the pages we released. */
50 	wait_queue_head_t acked;
51 
52 	/* Number of balloon pages we've told the Host we're not using. */
53 	unsigned int num_pages;
54 	/*
55 	 * The pages we've told the Host we're not using.
56 	 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
57 	 * to num_pages above.
58 	 */
59 	struct list_head pages;
60 
61 	/* The array of pfns we tell the Host about. */
62 	unsigned int num_pfns;
63 	u32 pfns[256];
64 
65 	/* Memory statistics */
66 	int need_stats_update;
67 	struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
68 };
69 
70 static struct virtio_device_id id_table[] = {
71 	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
72 	{ 0 },
73 };
74 
75 static u32 page_to_balloon_pfn(struct page *page)
76 {
77 	unsigned long pfn = page_to_pfn(page);
78 
79 	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
80 	/* Convert pfn from Linux page size to balloon page size. */
81 	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
82 }
83 
84 static struct page *balloon_pfn_to_page(u32 pfn)
85 {
86 	BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE);
87 	return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE);
88 }
89 
90 static void balloon_ack(struct virtqueue *vq)
91 {
92 	struct virtio_balloon *vb = vq->vdev->priv;
93 
94 	wake_up(&vb->acked);
95 }
96 
97 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
98 {
99 	struct scatterlist sg;
100 	unsigned int len;
101 
102 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
103 
104 	/* We should always be able to add one buffer to an empty queue. */
105 	if (virtqueue_add_buf(vq, &sg, 1, 0, vb, GFP_KERNEL) < 0)
106 		BUG();
107 	virtqueue_kick(vq);
108 
109 	/* When host has read buffer, this completes via balloon_ack */
110 	wait_event(vb->acked, virtqueue_get_buf(vq, &len));
111 }
112 
113 static void set_page_pfns(u32 pfns[], struct page *page)
114 {
115 	unsigned int i;
116 
117 	/* Set balloon pfns pointing at this page.
118 	 * Note that the first pfn points at start of the page. */
119 	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
120 		pfns[i] = page_to_balloon_pfn(page) + i;
121 }
122 
123 static void fill_balloon(struct virtio_balloon *vb, size_t num)
124 {
125 	/* We can only do one array worth at a time. */
126 	num = min(num, ARRAY_SIZE(vb->pfns));
127 
128 	for (vb->num_pfns = 0; vb->num_pfns < num;
129 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
130 		struct page *page = alloc_page(GFP_HIGHUSER | __GFP_NORETRY |
131 					__GFP_NOMEMALLOC | __GFP_NOWARN);
132 		if (!page) {
133 			if (printk_ratelimit())
134 				dev_printk(KERN_INFO, &vb->vdev->dev,
135 					   "Out of puff! Can't get %zu pages\n",
136 					   num);
137 			/* Sleep for at least 1/5 of a second before retry. */
138 			msleep(200);
139 			break;
140 		}
141 		set_page_pfns(vb->pfns + vb->num_pfns, page);
142 		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
143 		totalram_pages--;
144 		list_add(&page->lru, &vb->pages);
145 	}
146 
147 	/* Didn't get any?  Oh well. */
148 	if (vb->num_pfns == 0)
149 		return;
150 
151 	tell_host(vb, vb->inflate_vq);
152 }
153 
154 static void release_pages_by_pfn(const u32 pfns[], unsigned int num)
155 {
156 	unsigned int i;
157 
158 	/* Find pfns pointing at start of each page, get pages and free them. */
159 	for (i = 0; i < num; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
160 		__free_page(balloon_pfn_to_page(pfns[i]));
161 		totalram_pages++;
162 	}
163 }
164 
165 static void leak_balloon(struct virtio_balloon *vb, size_t num)
166 {
167 	struct page *page;
168 
169 	/* We can only do one array worth at a time. */
170 	num = min(num, ARRAY_SIZE(vb->pfns));
171 
172 	for (vb->num_pfns = 0; vb->num_pfns < num;
173 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
174 		page = list_first_entry(&vb->pages, struct page, lru);
175 		list_del(&page->lru);
176 		set_page_pfns(vb->pfns + vb->num_pfns, page);
177 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
178 	}
179 
180 	/*
181 	 * Note that if
182 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
183 	 * is true, we *have* to do it in this order
184 	 */
185 	tell_host(vb, vb->deflate_vq);
186 	release_pages_by_pfn(vb->pfns, vb->num_pfns);
187 }
188 
189 static inline void update_stat(struct virtio_balloon *vb, int idx,
190 			       u16 tag, u64 val)
191 {
192 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
193 	vb->stats[idx].tag = tag;
194 	vb->stats[idx].val = val;
195 }
196 
197 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
198 
199 static void update_balloon_stats(struct virtio_balloon *vb)
200 {
201 	unsigned long events[NR_VM_EVENT_ITEMS];
202 	struct sysinfo i;
203 	int idx = 0;
204 
205 	all_vm_events(events);
206 	si_meminfo(&i);
207 
208 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
209 				pages_to_bytes(events[PSWPIN]));
210 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
211 				pages_to_bytes(events[PSWPOUT]));
212 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
213 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
214 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
215 				pages_to_bytes(i.freeram));
216 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
217 				pages_to_bytes(i.totalram));
218 }
219 
220 /*
221  * While most virtqueues communicate guest-initiated requests to the hypervisor,
222  * the stats queue operates in reverse.  The driver initializes the virtqueue
223  * with a single buffer.  From that point forward, all conversations consist of
224  * a hypervisor request (a call to this function) which directs us to refill
225  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
226  * we notify our kthread which does the actual work via stats_handle_request().
227  */
228 static void stats_request(struct virtqueue *vq)
229 {
230 	struct virtio_balloon *vb = vq->vdev->priv;
231 
232 	vb->need_stats_update = 1;
233 	wake_up(&vb->config_change);
234 }
235 
236 static void stats_handle_request(struct virtio_balloon *vb)
237 {
238 	struct virtqueue *vq;
239 	struct scatterlist sg;
240 	unsigned int len;
241 
242 	vb->need_stats_update = 0;
243 	update_balloon_stats(vb);
244 
245 	vq = vb->stats_vq;
246 	if (!virtqueue_get_buf(vq, &len))
247 		return;
248 	sg_init_one(&sg, vb->stats, sizeof(vb->stats));
249 	if (virtqueue_add_buf(vq, &sg, 1, 0, vb, GFP_KERNEL) < 0)
250 		BUG();
251 	virtqueue_kick(vq);
252 }
253 
254 static void virtballoon_changed(struct virtio_device *vdev)
255 {
256 	struct virtio_balloon *vb = vdev->priv;
257 
258 	wake_up(&vb->config_change);
259 }
260 
261 static inline s64 towards_target(struct virtio_balloon *vb)
262 {
263 	__le32 v;
264 	s64 target;
265 
266 	vb->vdev->config->get(vb->vdev,
267 			      offsetof(struct virtio_balloon_config, num_pages),
268 			      &v, sizeof(v));
269 	target = le32_to_cpu(v);
270 	return target - vb->num_pages;
271 }
272 
273 static void update_balloon_size(struct virtio_balloon *vb)
274 {
275 	__le32 actual = cpu_to_le32(vb->num_pages);
276 
277 	vb->vdev->config->set(vb->vdev,
278 			      offsetof(struct virtio_balloon_config, actual),
279 			      &actual, sizeof(actual));
280 }
281 
282 static int balloon(void *_vballoon)
283 {
284 	struct virtio_balloon *vb = _vballoon;
285 
286 	set_freezable();
287 	while (!kthread_should_stop()) {
288 		s64 diff;
289 
290 		try_to_freeze();
291 		wait_event_interruptible(vb->config_change,
292 					 (diff = towards_target(vb)) != 0
293 					 || vb->need_stats_update
294 					 || kthread_should_stop()
295 					 || freezing(current));
296 		if (vb->need_stats_update)
297 			stats_handle_request(vb);
298 		if (diff > 0)
299 			fill_balloon(vb, diff);
300 		else if (diff < 0)
301 			leak_balloon(vb, -diff);
302 		update_balloon_size(vb);
303 	}
304 	return 0;
305 }
306 
307 static int init_vqs(struct virtio_balloon *vb)
308 {
309 	struct virtqueue *vqs[3];
310 	vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
311 	const char *names[] = { "inflate", "deflate", "stats" };
312 	int err, nvqs;
313 
314 	/*
315 	 * We expect two virtqueues: inflate and deflate, and
316 	 * optionally stat.
317 	 */
318 	nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
319 	err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
320 	if (err)
321 		return err;
322 
323 	vb->inflate_vq = vqs[0];
324 	vb->deflate_vq = vqs[1];
325 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
326 		struct scatterlist sg;
327 		vb->stats_vq = vqs[2];
328 
329 		/*
330 		 * Prime this virtqueue with one buffer so the hypervisor can
331 		 * use it to signal us later.
332 		 */
333 		sg_init_one(&sg, vb->stats, sizeof vb->stats);
334 		if (virtqueue_add_buf(vb->stats_vq, &sg, 1, 0, vb, GFP_KERNEL)
335 		    < 0)
336 			BUG();
337 		virtqueue_kick(vb->stats_vq);
338 	}
339 	return 0;
340 }
341 
342 static int virtballoon_probe(struct virtio_device *vdev)
343 {
344 	struct virtio_balloon *vb;
345 	int err;
346 
347 	vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
348 	if (!vb) {
349 		err = -ENOMEM;
350 		goto out;
351 	}
352 
353 	INIT_LIST_HEAD(&vb->pages);
354 	vb->num_pages = 0;
355 	init_waitqueue_head(&vb->config_change);
356 	init_waitqueue_head(&vb->acked);
357 	vb->vdev = vdev;
358 	vb->need_stats_update = 0;
359 
360 	err = init_vqs(vb);
361 	if (err)
362 		goto out_free_vb;
363 
364 	vb->thread = kthread_run(balloon, vb, "vballoon");
365 	if (IS_ERR(vb->thread)) {
366 		err = PTR_ERR(vb->thread);
367 		goto out_del_vqs;
368 	}
369 
370 	return 0;
371 
372 out_del_vqs:
373 	vdev->config->del_vqs(vdev);
374 out_free_vb:
375 	kfree(vb);
376 out:
377 	return err;
378 }
379 
380 static void remove_common(struct virtio_balloon *vb)
381 {
382 	/* There might be pages left in the balloon: free them. */
383 	while (vb->num_pages)
384 		leak_balloon(vb, vb->num_pages);
385 	update_balloon_size(vb);
386 
387 	/* Now we reset the device so we can clean up the queues. */
388 	vb->vdev->config->reset(vb->vdev);
389 
390 	vb->vdev->config->del_vqs(vb->vdev);
391 }
392 
393 static void __devexit virtballoon_remove(struct virtio_device *vdev)
394 {
395 	struct virtio_balloon *vb = vdev->priv;
396 
397 	kthread_stop(vb->thread);
398 	remove_common(vb);
399 	kfree(vb);
400 }
401 
402 #ifdef CONFIG_PM
403 static int virtballoon_freeze(struct virtio_device *vdev)
404 {
405 	struct virtio_balloon *vb = vdev->priv;
406 
407 	/*
408 	 * The kthread is already frozen by the PM core before this
409 	 * function is called.
410 	 */
411 
412 	remove_common(vb);
413 	return 0;
414 }
415 
416 static int virtballoon_restore(struct virtio_device *vdev)
417 {
418 	struct virtio_balloon *vb = vdev->priv;
419 	int ret;
420 
421 	ret = init_vqs(vdev->priv);
422 	if (ret)
423 		return ret;
424 
425 	fill_balloon(vb, towards_target(vb));
426 	update_balloon_size(vb);
427 	return 0;
428 }
429 #endif
430 
431 static unsigned int features[] = {
432 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
433 	VIRTIO_BALLOON_F_STATS_VQ,
434 };
435 
436 static struct virtio_driver virtio_balloon_driver = {
437 	.feature_table = features,
438 	.feature_table_size = ARRAY_SIZE(features),
439 	.driver.name =	KBUILD_MODNAME,
440 	.driver.owner =	THIS_MODULE,
441 	.id_table =	id_table,
442 	.probe =	virtballoon_probe,
443 	.remove =	__devexit_p(virtballoon_remove),
444 	.config_changed = virtballoon_changed,
445 #ifdef CONFIG_PM
446 	.freeze	=	virtballoon_freeze,
447 	.restore =	virtballoon_restore,
448 #endif
449 };
450 
451 static int __init init(void)
452 {
453 	return register_virtio_driver(&virtio_balloon_driver);
454 }
455 
456 static void __exit fini(void)
457 {
458 	unregister_virtio_driver(&virtio_balloon_driver);
459 }
460 module_init(init);
461 module_exit(fini);
462 
463 MODULE_DEVICE_TABLE(virtio, id_table);
464 MODULE_DESCRIPTION("Virtio balloon driver");
465 MODULE_LICENSE("GPL");
466