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