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 	struct completion 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;
93 	unsigned int len;
94 
95 	vb = virtqueue_get_buf(vq, &len);
96 	if (vb)
97 		complete(&vb->acked);
98 }
99 
100 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
101 {
102 	struct scatterlist sg;
103 
104 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
105 
106 	init_completion(&vb->acked);
107 
108 	/* We should always be able to add one buffer to an empty queue. */
109 	if (virtqueue_add_buf(vq, &sg, 1, 0, vb, GFP_KERNEL) < 0)
110 		BUG();
111 	virtqueue_kick(vq);
112 
113 	/* When host has read buffer, this completes via balloon_ack */
114 	wait_for_completion(&vb->acked);
115 }
116 
117 static void set_page_pfns(u32 pfns[], struct page *page)
118 {
119 	unsigned int i;
120 
121 	/* Set balloon pfns pointing at this page.
122 	 * Note that the first pfn points at start of the page. */
123 	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
124 		pfns[i] = page_to_balloon_pfn(page) + i;
125 }
126 
127 static void fill_balloon(struct virtio_balloon *vb, size_t num)
128 {
129 	/* We can only do one array worth at a time. */
130 	num = min(num, ARRAY_SIZE(vb->pfns));
131 
132 	for (vb->num_pfns = 0; vb->num_pfns < num;
133 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
134 		struct page *page = alloc_page(GFP_HIGHUSER | __GFP_NORETRY |
135 					__GFP_NOMEMALLOC | __GFP_NOWARN);
136 		if (!page) {
137 			if (printk_ratelimit())
138 				dev_printk(KERN_INFO, &vb->vdev->dev,
139 					   "Out of puff! Can't get %zu pages\n",
140 					   num);
141 			/* Sleep for at least 1/5 of a second before retry. */
142 			msleep(200);
143 			break;
144 		}
145 		set_page_pfns(vb->pfns + vb->num_pfns, page);
146 		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
147 		totalram_pages--;
148 		list_add(&page->lru, &vb->pages);
149 	}
150 
151 	/* Didn't get any?  Oh well. */
152 	if (vb->num_pfns == 0)
153 		return;
154 
155 	tell_host(vb, vb->inflate_vq);
156 }
157 
158 static void release_pages_by_pfn(const u32 pfns[], unsigned int num)
159 {
160 	unsigned int i;
161 
162 	/* Find pfns pointing at start of each page, get pages and free them. */
163 	for (i = 0; i < num; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
164 		__free_page(balloon_pfn_to_page(pfns[i]));
165 		totalram_pages++;
166 	}
167 }
168 
169 static void leak_balloon(struct virtio_balloon *vb, size_t num)
170 {
171 	struct page *page;
172 
173 	/* We can only do one array worth at a time. */
174 	num = min(num, ARRAY_SIZE(vb->pfns));
175 
176 	for (vb->num_pfns = 0; vb->num_pfns < num;
177 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
178 		page = list_first_entry(&vb->pages, struct page, lru);
179 		list_del(&page->lru);
180 		set_page_pfns(vb->pfns + vb->num_pfns, page);
181 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
182 	}
183 
184 	/*
185 	 * Note that if
186 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
187 	 * is true, we *have* to do it in this order
188 	 */
189 	tell_host(vb, vb->deflate_vq);
190 	release_pages_by_pfn(vb->pfns, vb->num_pfns);
191 }
192 
193 static inline void update_stat(struct virtio_balloon *vb, int idx,
194 			       u16 tag, u64 val)
195 {
196 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
197 	vb->stats[idx].tag = tag;
198 	vb->stats[idx].val = val;
199 }
200 
201 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
202 
203 static void update_balloon_stats(struct virtio_balloon *vb)
204 {
205 	unsigned long events[NR_VM_EVENT_ITEMS];
206 	struct sysinfo i;
207 	int idx = 0;
208 
209 	all_vm_events(events);
210 	si_meminfo(&i);
211 
212 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
213 				pages_to_bytes(events[PSWPIN]));
214 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
215 				pages_to_bytes(events[PSWPOUT]));
216 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
217 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
218 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
219 				pages_to_bytes(i.freeram));
220 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
221 				pages_to_bytes(i.totalram));
222 }
223 
224 /*
225  * While most virtqueues communicate guest-initiated requests to the hypervisor,
226  * the stats queue operates in reverse.  The driver initializes the virtqueue
227  * with a single buffer.  From that point forward, all conversations consist of
228  * a hypervisor request (a call to this function) which directs us to refill
229  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
230  * we notify our kthread which does the actual work via stats_handle_request().
231  */
232 static void stats_request(struct virtqueue *vq)
233 {
234 	struct virtio_balloon *vb;
235 	unsigned int len;
236 
237 	vb = virtqueue_get_buf(vq, &len);
238 	if (!vb)
239 		return;
240 	vb->need_stats_update = 1;
241 	wake_up(&vb->config_change);
242 }
243 
244 static void stats_handle_request(struct virtio_balloon *vb)
245 {
246 	struct virtqueue *vq;
247 	struct scatterlist sg;
248 
249 	vb->need_stats_update = 0;
250 	update_balloon_stats(vb);
251 
252 	vq = vb->stats_vq;
253 	sg_init_one(&sg, vb->stats, sizeof(vb->stats));
254 	if (virtqueue_add_buf(vq, &sg, 1, 0, vb, GFP_KERNEL) < 0)
255 		BUG();
256 	virtqueue_kick(vq);
257 }
258 
259 static void virtballoon_changed(struct virtio_device *vdev)
260 {
261 	struct virtio_balloon *vb = vdev->priv;
262 
263 	wake_up(&vb->config_change);
264 }
265 
266 static inline s64 towards_target(struct virtio_balloon *vb)
267 {
268 	__le32 v;
269 	s64 target;
270 
271 	vb->vdev->config->get(vb->vdev,
272 			      offsetof(struct virtio_balloon_config, num_pages),
273 			      &v, sizeof(v));
274 	target = le32_to_cpu(v);
275 	return target - vb->num_pages;
276 }
277 
278 static void update_balloon_size(struct virtio_balloon *vb)
279 {
280 	__le32 actual = cpu_to_le32(vb->num_pages);
281 
282 	vb->vdev->config->set(vb->vdev,
283 			      offsetof(struct virtio_balloon_config, actual),
284 			      &actual, sizeof(actual));
285 }
286 
287 static int balloon(void *_vballoon)
288 {
289 	struct virtio_balloon *vb = _vballoon;
290 
291 	set_freezable();
292 	while (!kthread_should_stop()) {
293 		s64 diff;
294 
295 		try_to_freeze();
296 		wait_event_interruptible(vb->config_change,
297 					 (diff = towards_target(vb)) != 0
298 					 || vb->need_stats_update
299 					 || kthread_should_stop()
300 					 || freezing(current));
301 		if (vb->need_stats_update)
302 			stats_handle_request(vb);
303 		if (diff > 0)
304 			fill_balloon(vb, diff);
305 		else if (diff < 0)
306 			leak_balloon(vb, -diff);
307 		update_balloon_size(vb);
308 	}
309 	return 0;
310 }
311 
312 static int init_vqs(struct virtio_balloon *vb)
313 {
314 	struct virtqueue *vqs[3];
315 	vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
316 	const char *names[] = { "inflate", "deflate", "stats" };
317 	int err, nvqs;
318 
319 	/*
320 	 * We expect two virtqueues: inflate and deflate, and
321 	 * optionally stat.
322 	 */
323 	nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
324 	err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
325 	if (err)
326 		return err;
327 
328 	vb->inflate_vq = vqs[0];
329 	vb->deflate_vq = vqs[1];
330 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
331 		struct scatterlist sg;
332 		vb->stats_vq = vqs[2];
333 
334 		/*
335 		 * Prime this virtqueue with one buffer so the hypervisor can
336 		 * use it to signal us later.
337 		 */
338 		sg_init_one(&sg, vb->stats, sizeof vb->stats);
339 		if (virtqueue_add_buf(vb->stats_vq, &sg, 1, 0, vb, GFP_KERNEL)
340 		    < 0)
341 			BUG();
342 		virtqueue_kick(vb->stats_vq);
343 	}
344 	return 0;
345 }
346 
347 static int virtballoon_probe(struct virtio_device *vdev)
348 {
349 	struct virtio_balloon *vb;
350 	int err;
351 
352 	vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
353 	if (!vb) {
354 		err = -ENOMEM;
355 		goto out;
356 	}
357 
358 	INIT_LIST_HEAD(&vb->pages);
359 	vb->num_pages = 0;
360 	init_waitqueue_head(&vb->config_change);
361 	vb->vdev = vdev;
362 	vb->need_stats_update = 0;
363 
364 	err = init_vqs(vb);
365 	if (err)
366 		goto out_free_vb;
367 
368 	vb->thread = kthread_run(balloon, vb, "vballoon");
369 	if (IS_ERR(vb->thread)) {
370 		err = PTR_ERR(vb->thread);
371 		goto out_del_vqs;
372 	}
373 
374 	return 0;
375 
376 out_del_vqs:
377 	vdev->config->del_vqs(vdev);
378 out_free_vb:
379 	kfree(vb);
380 out:
381 	return err;
382 }
383 
384 static void __devexit virtballoon_remove(struct virtio_device *vdev)
385 {
386 	struct virtio_balloon *vb = vdev->priv;
387 
388 	kthread_stop(vb->thread);
389 
390 	/* There might be pages left in the balloon: free them. */
391 	while (vb->num_pages)
392 		leak_balloon(vb, vb->num_pages);
393 	update_balloon_size(vb);
394 
395 	/* Now we reset the device so we can clean up the queues. */
396 	vdev->config->reset(vdev);
397 
398 	vdev->config->del_vqs(vdev);
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 	while (vb->num_pages)
413 		leak_balloon(vb, vb->num_pages);
414 	update_balloon_size(vb);
415 
416 	/* Ensure we don't get any more requests from the host */
417 	vdev->config->reset(vdev);
418 	vdev->config->del_vqs(vdev);
419 	return 0;
420 }
421 
422 static int restore_common(struct virtio_device *vdev)
423 {
424 	struct virtio_balloon *vb = vdev->priv;
425 	int ret;
426 
427 	ret = init_vqs(vdev->priv);
428 	if (ret)
429 		return ret;
430 
431 	fill_balloon(vb, towards_target(vb));
432 	update_balloon_size(vb);
433 	return 0;
434 }
435 
436 static int virtballoon_restore(struct virtio_device *vdev)
437 {
438 	return restore_common(vdev);
439 }
440 #endif
441 
442 static unsigned int features[] = {
443 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
444 	VIRTIO_BALLOON_F_STATS_VQ,
445 };
446 
447 static struct virtio_driver virtio_balloon_driver = {
448 	.feature_table = features,
449 	.feature_table_size = ARRAY_SIZE(features),
450 	.driver.name =	KBUILD_MODNAME,
451 	.driver.owner =	THIS_MODULE,
452 	.id_table =	id_table,
453 	.probe =	virtballoon_probe,
454 	.remove =	__devexit_p(virtballoon_remove),
455 	.config_changed = virtballoon_changed,
456 #ifdef CONFIG_PM
457 	.freeze	=	virtballoon_freeze,
458 	.restore =	virtballoon_restore,
459 #endif
460 };
461 
462 static int __init init(void)
463 {
464 	return register_virtio_driver(&virtio_balloon_driver);
465 }
466 
467 static void __exit fini(void)
468 {
469 	unregister_virtio_driver(&virtio_balloon_driver);
470 }
471 module_init(init);
472 module_exit(fini);
473 
474 MODULE_DEVICE_TABLE(virtio, id_table);
475 MODULE_DESCRIPTION("Virtio balloon driver");
476 MODULE_LICENSE("GPL");
477