Commit | Line | Data |
---|---|---|
6b35e407 RR |
1 | /* Virtio balloon implementation, inspired by Dor Loar and Marcelo |
2 | * Tosatti's implementations. | |
3 | * | |
4 | * Copyright 2008 Rusty Russell IBM Corporation | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program; if not, write to the Free Software | |
18 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
19 | */ | |
20 | //#define DEBUG | |
21 | #include <linux/virtio.h> | |
22 | #include <linux/virtio_balloon.h> | |
23 | #include <linux/swap.h> | |
24 | #include <linux/kthread.h> | |
25 | #include <linux/freezer.h> | |
6659a0f0 | 26 | #include <linux/delay.h> |
6b35e407 RR |
27 | |
28 | struct virtio_balloon | |
29 | { | |
30 | struct virtio_device *vdev; | |
9564e138 | 31 | struct virtqueue *inflate_vq, *deflate_vq, *stats_vq; |
6b35e407 RR |
32 | |
33 | /* Where the ballooning thread waits for config to change. */ | |
34 | wait_queue_head_t config_change; | |
35 | ||
36 | /* The thread servicing the balloon. */ | |
37 | struct task_struct *thread; | |
38 | ||
39 | /* Waiting for host to ack the pages we released. */ | |
40 | struct completion acked; | |
41 | ||
42 | /* Do we have to tell Host *before* we reuse pages? */ | |
43 | bool tell_host_first; | |
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]; | |
9564e138 AL |
52 | |
53 | /* Memory statistics */ | |
54 | struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR]; | |
6b35e407 RR |
55 | }; |
56 | ||
57 | static struct virtio_device_id id_table[] = { | |
58 | { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID }, | |
59 | { 0 }, | |
60 | }; | |
61 | ||
1b4aa2fa HB |
62 | static u32 page_to_balloon_pfn(struct page *page) |
63 | { | |
64 | unsigned long pfn = page_to_pfn(page); | |
65 | ||
66 | BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT); | |
67 | /* Convert pfn from Linux page size to balloon page size. */ | |
68 | return pfn >> (PAGE_SHIFT - VIRTIO_BALLOON_PFN_SHIFT); | |
69 | } | |
70 | ||
6b35e407 RR |
71 | static void balloon_ack(struct virtqueue *vq) |
72 | { | |
73 | struct virtio_balloon *vb; | |
74 | unsigned int len; | |
75 | ||
76 | vb = vq->vq_ops->get_buf(vq, &len); | |
77 | if (vb) | |
78 | complete(&vb->acked); | |
79 | } | |
80 | ||
81 | static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq) | |
82 | { | |
83 | struct scatterlist sg; | |
84 | ||
85 | sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns); | |
86 | ||
87 | init_completion(&vb->acked); | |
88 | ||
89 | /* We should always be able to add one buffer to an empty queue. */ | |
3c1b27d5 | 90 | if (vq->vq_ops->add_buf(vq, &sg, 1, 0, vb) < 0) |
6b35e407 RR |
91 | BUG(); |
92 | vq->vq_ops->kick(vq); | |
93 | ||
94 | /* When host has read buffer, this completes via balloon_ack */ | |
95 | wait_for_completion(&vb->acked); | |
96 | } | |
97 | ||
98 | static void fill_balloon(struct virtio_balloon *vb, size_t num) | |
99 | { | |
100 | /* We can only do one array worth at a time. */ | |
101 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
102 | ||
103 | for (vb->num_pfns = 0; vb->num_pfns < num; vb->num_pfns++) { | |
104 | struct page *page = alloc_page(GFP_HIGHUSER | __GFP_NORETRY); | |
105 | if (!page) { | |
106 | if (printk_ratelimit()) | |
107 | dev_printk(KERN_INFO, &vb->vdev->dev, | |
108 | "Out of puff! Can't get %zu pages\n", | |
109 | num); | |
110 | /* Sleep for at least 1/5 of a second before retry. */ | |
111 | msleep(200); | |
112 | break; | |
113 | } | |
1b4aa2fa | 114 | vb->pfns[vb->num_pfns] = page_to_balloon_pfn(page); |
6b35e407 RR |
115 | totalram_pages--; |
116 | vb->num_pages++; | |
117 | list_add(&page->lru, &vb->pages); | |
118 | } | |
119 | ||
120 | /* Didn't get any? Oh well. */ | |
121 | if (vb->num_pfns == 0) | |
122 | return; | |
123 | ||
124 | tell_host(vb, vb->inflate_vq); | |
125 | } | |
126 | ||
127 | static void release_pages_by_pfn(const u32 pfns[], unsigned int num) | |
128 | { | |
129 | unsigned int i; | |
130 | ||
131 | for (i = 0; i < num; i++) { | |
132 | __free_page(pfn_to_page(pfns[i])); | |
133 | totalram_pages++; | |
134 | } | |
135 | } | |
136 | ||
137 | static void leak_balloon(struct virtio_balloon *vb, size_t num) | |
138 | { | |
139 | struct page *page; | |
140 | ||
141 | /* We can only do one array worth at a time. */ | |
142 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
143 | ||
144 | for (vb->num_pfns = 0; vb->num_pfns < num; vb->num_pfns++) { | |
145 | page = list_first_entry(&vb->pages, struct page, lru); | |
146 | list_del(&page->lru); | |
1b4aa2fa | 147 | vb->pfns[vb->num_pfns] = page_to_balloon_pfn(page); |
6b35e407 RR |
148 | vb->num_pages--; |
149 | } | |
150 | ||
151 | if (vb->tell_host_first) { | |
152 | tell_host(vb, vb->deflate_vq); | |
153 | release_pages_by_pfn(vb->pfns, vb->num_pfns); | |
154 | } else { | |
155 | release_pages_by_pfn(vb->pfns, vb->num_pfns); | |
156 | tell_host(vb, vb->deflate_vq); | |
157 | } | |
158 | } | |
159 | ||
9564e138 AL |
160 | static inline void update_stat(struct virtio_balloon *vb, int idx, |
161 | u16 tag, u64 val) | |
162 | { | |
163 | BUG_ON(idx >= VIRTIO_BALLOON_S_NR); | |
164 | vb->stats[idx].tag = tag; | |
165 | vb->stats[idx].val = val; | |
166 | } | |
167 | ||
168 | #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT) | |
169 | ||
170 | static void update_balloon_stats(struct virtio_balloon *vb) | |
171 | { | |
172 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
173 | struct sysinfo i; | |
174 | int idx = 0; | |
175 | ||
176 | all_vm_events(events); | |
177 | si_meminfo(&i); | |
178 | ||
179 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN, | |
180 | pages_to_bytes(events[PSWPIN])); | |
181 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT, | |
182 | pages_to_bytes(events[PSWPOUT])); | |
183 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]); | |
184 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]); | |
185 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE, | |
186 | pages_to_bytes(i.freeram)); | |
187 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT, | |
188 | pages_to_bytes(i.totalram)); | |
189 | } | |
190 | ||
191 | /* | |
192 | * While most virtqueues communicate guest-initiated requests to the hypervisor, | |
193 | * the stats queue operates in reverse. The driver initializes the virtqueue | |
194 | * with a single buffer. From that point forward, all conversations consist of | |
195 | * a hypervisor request (a call to this function) which directs us to refill | |
196 | * the virtqueue with a fresh stats buffer. | |
197 | */ | |
198 | static void stats_ack(struct virtqueue *vq) | |
199 | { | |
200 | struct virtio_balloon *vb; | |
201 | unsigned int len; | |
202 | struct scatterlist sg; | |
203 | ||
204 | vb = vq->vq_ops->get_buf(vq, &len); | |
205 | if (!vb) | |
206 | return; | |
207 | ||
208 | update_balloon_stats(vb); | |
209 | ||
210 | sg_init_one(&sg, vb->stats, sizeof(vb->stats)); | |
211 | if (vq->vq_ops->add_buf(vq, &sg, 1, 0, vb) < 0) | |
212 | BUG(); | |
213 | vq->vq_ops->kick(vq); | |
214 | } | |
215 | ||
6b35e407 RR |
216 | static void virtballoon_changed(struct virtio_device *vdev) |
217 | { | |
218 | struct virtio_balloon *vb = vdev->priv; | |
219 | ||
220 | wake_up(&vb->config_change); | |
221 | } | |
222 | ||
bdc1681c | 223 | static inline s64 towards_target(struct virtio_balloon *vb) |
6b35e407 RR |
224 | { |
225 | u32 v; | |
72e61eb4 RR |
226 | vb->vdev->config->get(vb->vdev, |
227 | offsetof(struct virtio_balloon_config, num_pages), | |
228 | &v, sizeof(v)); | |
532a6086 | 229 | return (s64)v - vb->num_pages; |
6b35e407 RR |
230 | } |
231 | ||
232 | static void update_balloon_size(struct virtio_balloon *vb) | |
233 | { | |
234 | __le32 actual = cpu_to_le32(vb->num_pages); | |
235 | ||
236 | vb->vdev->config->set(vb->vdev, | |
237 | offsetof(struct virtio_balloon_config, actual), | |
238 | &actual, sizeof(actual)); | |
239 | } | |
240 | ||
241 | static int balloon(void *_vballoon) | |
242 | { | |
243 | struct virtio_balloon *vb = _vballoon; | |
244 | ||
245 | set_freezable(); | |
246 | while (!kthread_should_stop()) { | |
bdc1681c | 247 | s64 diff; |
6b35e407 RR |
248 | |
249 | try_to_freeze(); | |
250 | wait_event_interruptible(vb->config_change, | |
251 | (diff = towards_target(vb)) != 0 | |
84a139a9 MT |
252 | || kthread_should_stop() |
253 | || freezing(current)); | |
6b35e407 RR |
254 | if (diff > 0) |
255 | fill_balloon(vb, diff); | |
256 | else if (diff < 0) | |
257 | leak_balloon(vb, -diff); | |
258 | update_balloon_size(vb); | |
259 | } | |
260 | return 0; | |
261 | } | |
262 | ||
263 | static int virtballoon_probe(struct virtio_device *vdev) | |
264 | { | |
265 | struct virtio_balloon *vb; | |
9564e138 AL |
266 | struct virtqueue *vqs[3]; |
267 | vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_ack }; | |
268 | const char *names[] = { "inflate", "deflate", "stats" }; | |
269 | int err, nvqs; | |
6b35e407 RR |
270 | |
271 | vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL); | |
272 | if (!vb) { | |
273 | err = -ENOMEM; | |
274 | goto out; | |
275 | } | |
276 | ||
277 | INIT_LIST_HEAD(&vb->pages); | |
278 | vb->num_pages = 0; | |
279 | init_waitqueue_head(&vb->config_change); | |
280 | vb->vdev = vdev; | |
281 | ||
9564e138 AL |
282 | /* We expect two virtqueues: inflate and deflate, |
283 | * and optionally stat. */ | |
284 | nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2; | |
285 | err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names); | |
d2a7ddda | 286 | if (err) |
6b35e407 | 287 | goto out_free_vb; |
6b35e407 | 288 | |
d2a7ddda MT |
289 | vb->inflate_vq = vqs[0]; |
290 | vb->deflate_vq = vqs[1]; | |
9564e138 AL |
291 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { |
292 | struct scatterlist sg; | |
293 | vb->stats_vq = vqs[2]; | |
294 | ||
295 | /* | |
296 | * Prime this virtqueue with one buffer so the hypervisor can | |
297 | * use it to signal us later. | |
298 | */ | |
299 | sg_init_one(&sg, vb->stats, sizeof vb->stats); | |
300 | if (vb->stats_vq->vq_ops->add_buf(vb->stats_vq, | |
301 | &sg, 1, 0, vb) < 0) | |
302 | BUG(); | |
303 | vb->stats_vq->vq_ops->kick(vb->stats_vq); | |
304 | } | |
6b35e407 RR |
305 | |
306 | vb->thread = kthread_run(balloon, vb, "vballoon"); | |
307 | if (IS_ERR(vb->thread)) { | |
308 | err = PTR_ERR(vb->thread); | |
d2a7ddda | 309 | goto out_del_vqs; |
6b35e407 RR |
310 | } |
311 | ||
312 | vb->tell_host_first | |
c45a6816 | 313 | = virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST); |
6b35e407 RR |
314 | |
315 | return 0; | |
316 | ||
d2a7ddda MT |
317 | out_del_vqs: |
318 | vdev->config->del_vqs(vdev); | |
6b35e407 RR |
319 | out_free_vb: |
320 | kfree(vb); | |
321 | out: | |
322 | return err; | |
323 | } | |
324 | ||
1e65175c | 325 | static void __devexit virtballoon_remove(struct virtio_device *vdev) |
6b35e407 RR |
326 | { |
327 | struct virtio_balloon *vb = vdev->priv; | |
328 | ||
329 | kthread_stop(vb->thread); | |
330 | ||
331 | /* There might be pages left in the balloon: free them. */ | |
332 | while (vb->num_pages) | |
333 | leak_balloon(vb, vb->num_pages); | |
334 | ||
335 | /* Now we reset the device so we can clean up the queues. */ | |
336 | vdev->config->reset(vdev); | |
337 | ||
d2a7ddda | 338 | vdev->config->del_vqs(vdev); |
6b35e407 RR |
339 | kfree(vb); |
340 | } | |
341 | ||
9564e138 AL |
342 | static unsigned int features[] = { |
343 | VIRTIO_BALLOON_F_MUST_TELL_HOST, | |
344 | VIRTIO_BALLOON_F_STATS_VQ, | |
345 | }; | |
c45a6816 | 346 | |
d817cd52 | 347 | static struct virtio_driver virtio_balloon_driver = { |
c45a6816 RR |
348 | .feature_table = features, |
349 | .feature_table_size = ARRAY_SIZE(features), | |
6b35e407 RR |
350 | .driver.name = KBUILD_MODNAME, |
351 | .driver.owner = THIS_MODULE, | |
352 | .id_table = id_table, | |
353 | .probe = virtballoon_probe, | |
354 | .remove = __devexit_p(virtballoon_remove), | |
355 | .config_changed = virtballoon_changed, | |
356 | }; | |
357 | ||
358 | static int __init init(void) | |
359 | { | |
d817cd52 | 360 | return register_virtio_driver(&virtio_balloon_driver); |
6b35e407 RR |
361 | } |
362 | ||
363 | static void __exit fini(void) | |
364 | { | |
d817cd52 | 365 | unregister_virtio_driver(&virtio_balloon_driver); |
6b35e407 RR |
366 | } |
367 | module_init(init); | |
368 | module_exit(fini); | |
369 | ||
370 | MODULE_DEVICE_TABLE(virtio, id_table); | |
371 | MODULE_DESCRIPTION("Virtio balloon driver"); | |
372 | MODULE_LICENSE("GPL"); |