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> |
5a0e3ad6 | 27 | #include <linux/slab.h> |
6b35e407 RR |
28 | |
29 | struct virtio_balloon | |
30 | { | |
31 | struct virtio_device *vdev; | |
9564e138 | 32 | struct virtqueue *inflate_vq, *deflate_vq, *stats_vq; |
6b35e407 RR |
33 | |
34 | /* Where the ballooning thread waits for config to change. */ | |
35 | wait_queue_head_t config_change; | |
36 | ||
37 | /* The thread servicing the balloon. */ | |
38 | struct task_struct *thread; | |
39 | ||
40 | /* Waiting for host to ack the pages we released. */ | |
41 | struct completion acked; | |
42 | ||
6b35e407 RR |
43 | /* The pages we've told the Host we're not using. */ |
44 | unsigned int num_pages; | |
45 | struct list_head pages; | |
46 | ||
47 | /* The array of pfns we tell the Host about. */ | |
48 | unsigned int num_pfns; | |
49 | u32 pfns[256]; | |
9564e138 AL |
50 | |
51 | /* Memory statistics */ | |
1f34c71a | 52 | int need_stats_update; |
9564e138 | 53 | struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR]; |
6b35e407 RR |
54 | }; |
55 | ||
56 | static struct virtio_device_id id_table[] = { | |
57 | { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID }, | |
58 | { 0 }, | |
59 | }; | |
60 | ||
1b4aa2fa HB |
61 | static u32 page_to_balloon_pfn(struct page *page) |
62 | { | |
63 | unsigned long pfn = page_to_pfn(page); | |
64 | ||
65 | BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT); | |
66 | /* Convert pfn from Linux page size to balloon page size. */ | |
67 | return pfn >> (PAGE_SHIFT - VIRTIO_BALLOON_PFN_SHIFT); | |
68 | } | |
69 | ||
6b35e407 RR |
70 | static void balloon_ack(struct virtqueue *vq) |
71 | { | |
72 | struct virtio_balloon *vb; | |
73 | unsigned int len; | |
74 | ||
946cfe0e | 75 | vb = virtqueue_get_buf(vq, &len); |
6b35e407 RR |
76 | if (vb) |
77 | complete(&vb->acked); | |
78 | } | |
79 | ||
80 | static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq) | |
81 | { | |
82 | struct scatterlist sg; | |
83 | ||
84 | sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns); | |
85 | ||
86 | init_completion(&vb->acked); | |
87 | ||
88 | /* We should always be able to add one buffer to an empty queue. */ | |
946cfe0e | 89 | if (virtqueue_add_buf(vq, &sg, 1, 0, vb) < 0) |
6b35e407 | 90 | BUG(); |
946cfe0e | 91 | virtqueue_kick(vq); |
6b35e407 RR |
92 | |
93 | /* When host has read buffer, this completes via balloon_ack */ | |
94 | wait_for_completion(&vb->acked); | |
95 | } | |
96 | ||
97 | static void fill_balloon(struct virtio_balloon *vb, size_t num) | |
98 | { | |
99 | /* We can only do one array worth at a time. */ | |
100 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
101 | ||
102 | for (vb->num_pfns = 0; vb->num_pfns < num; vb->num_pfns++) { | |
61fb06cc BS |
103 | struct page *page = alloc_page(GFP_HIGHUSER | __GFP_NORETRY | |
104 | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
6b35e407 RR |
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 | ||
bf50e69f DH |
151 | |
152 | /* | |
153 | * Note that if | |
154 | * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST); | |
155 | * is true, we *have* to do it in this order | |
156 | */ | |
157 | tell_host(vb, vb->deflate_vq); | |
158 | release_pages_by_pfn(vb->pfns, vb->num_pfns); | |
6b35e407 RR |
159 | } |
160 | ||
9564e138 AL |
161 | static inline void update_stat(struct virtio_balloon *vb, int idx, |
162 | u16 tag, u64 val) | |
163 | { | |
164 | BUG_ON(idx >= VIRTIO_BALLOON_S_NR); | |
165 | vb->stats[idx].tag = tag; | |
166 | vb->stats[idx].val = val; | |
167 | } | |
168 | ||
169 | #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT) | |
170 | ||
171 | static void update_balloon_stats(struct virtio_balloon *vb) | |
172 | { | |
173 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
174 | struct sysinfo i; | |
175 | int idx = 0; | |
176 | ||
177 | all_vm_events(events); | |
178 | si_meminfo(&i); | |
179 | ||
180 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN, | |
181 | pages_to_bytes(events[PSWPIN])); | |
182 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT, | |
183 | pages_to_bytes(events[PSWPOUT])); | |
184 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]); | |
185 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]); | |
186 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE, | |
187 | pages_to_bytes(i.freeram)); | |
188 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT, | |
189 | pages_to_bytes(i.totalram)); | |
190 | } | |
191 | ||
192 | /* | |
193 | * While most virtqueues communicate guest-initiated requests to the hypervisor, | |
194 | * the stats queue operates in reverse. The driver initializes the virtqueue | |
195 | * with a single buffer. From that point forward, all conversations consist of | |
196 | * a hypervisor request (a call to this function) which directs us to refill | |
1f34c71a AL |
197 | * the virtqueue with a fresh stats buffer. Since stats collection can sleep, |
198 | * we notify our kthread which does the actual work via stats_handle_request(). | |
9564e138 | 199 | */ |
1f34c71a | 200 | static void stats_request(struct virtqueue *vq) |
9564e138 AL |
201 | { |
202 | struct virtio_balloon *vb; | |
203 | unsigned int len; | |
9564e138 | 204 | |
946cfe0e | 205 | vb = virtqueue_get_buf(vq, &len); |
9564e138 AL |
206 | if (!vb) |
207 | return; | |
1f34c71a AL |
208 | vb->need_stats_update = 1; |
209 | wake_up(&vb->config_change); | |
210 | } | |
211 | ||
212 | static void stats_handle_request(struct virtio_balloon *vb) | |
213 | { | |
214 | struct virtqueue *vq; | |
215 | struct scatterlist sg; | |
9564e138 | 216 | |
1f34c71a | 217 | vb->need_stats_update = 0; |
9564e138 AL |
218 | update_balloon_stats(vb); |
219 | ||
1f34c71a | 220 | vq = vb->stats_vq; |
9564e138 | 221 | sg_init_one(&sg, vb->stats, sizeof(vb->stats)); |
946cfe0e | 222 | if (virtqueue_add_buf(vq, &sg, 1, 0, vb) < 0) |
9564e138 | 223 | BUG(); |
946cfe0e | 224 | virtqueue_kick(vq); |
9564e138 AL |
225 | } |
226 | ||
6b35e407 RR |
227 | static void virtballoon_changed(struct virtio_device *vdev) |
228 | { | |
229 | struct virtio_balloon *vb = vdev->priv; | |
230 | ||
231 | wake_up(&vb->config_change); | |
232 | } | |
233 | ||
bdc1681c | 234 | static inline s64 towards_target(struct virtio_balloon *vb) |
6b35e407 RR |
235 | { |
236 | u32 v; | |
72e61eb4 RR |
237 | vb->vdev->config->get(vb->vdev, |
238 | offsetof(struct virtio_balloon_config, num_pages), | |
239 | &v, sizeof(v)); | |
532a6086 | 240 | return (s64)v - vb->num_pages; |
6b35e407 RR |
241 | } |
242 | ||
243 | static void update_balloon_size(struct virtio_balloon *vb) | |
244 | { | |
245 | __le32 actual = cpu_to_le32(vb->num_pages); | |
246 | ||
247 | vb->vdev->config->set(vb->vdev, | |
248 | offsetof(struct virtio_balloon_config, actual), | |
249 | &actual, sizeof(actual)); | |
250 | } | |
251 | ||
252 | static int balloon(void *_vballoon) | |
253 | { | |
254 | struct virtio_balloon *vb = _vballoon; | |
255 | ||
256 | set_freezable(); | |
257 | while (!kthread_should_stop()) { | |
bdc1681c | 258 | s64 diff; |
6b35e407 RR |
259 | |
260 | try_to_freeze(); | |
261 | wait_event_interruptible(vb->config_change, | |
262 | (diff = towards_target(vb)) != 0 | |
1f34c71a | 263 | || vb->need_stats_update |
84a139a9 MT |
264 | || kthread_should_stop() |
265 | || freezing(current)); | |
1f34c71a AL |
266 | if (vb->need_stats_update) |
267 | stats_handle_request(vb); | |
6b35e407 RR |
268 | if (diff > 0) |
269 | fill_balloon(vb, diff); | |
270 | else if (diff < 0) | |
271 | leak_balloon(vb, -diff); | |
272 | update_balloon_size(vb); | |
273 | } | |
274 | return 0; | |
275 | } | |
276 | ||
277 | static int virtballoon_probe(struct virtio_device *vdev) | |
278 | { | |
279 | struct virtio_balloon *vb; | |
9564e138 | 280 | struct virtqueue *vqs[3]; |
1f34c71a | 281 | vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request }; |
9564e138 AL |
282 | const char *names[] = { "inflate", "deflate", "stats" }; |
283 | int err, nvqs; | |
6b35e407 RR |
284 | |
285 | vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL); | |
286 | if (!vb) { | |
287 | err = -ENOMEM; | |
288 | goto out; | |
289 | } | |
290 | ||
291 | INIT_LIST_HEAD(&vb->pages); | |
292 | vb->num_pages = 0; | |
293 | init_waitqueue_head(&vb->config_change); | |
294 | vb->vdev = vdev; | |
169c246a | 295 | vb->need_stats_update = 0; |
6b35e407 | 296 | |
9564e138 AL |
297 | /* We expect two virtqueues: inflate and deflate, |
298 | * and optionally stat. */ | |
299 | nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2; | |
300 | err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names); | |
d2a7ddda | 301 | if (err) |
6b35e407 | 302 | goto out_free_vb; |
6b35e407 | 303 | |
d2a7ddda MT |
304 | vb->inflate_vq = vqs[0]; |
305 | vb->deflate_vq = vqs[1]; | |
9564e138 AL |
306 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { |
307 | struct scatterlist sg; | |
308 | vb->stats_vq = vqs[2]; | |
309 | ||
310 | /* | |
311 | * Prime this virtqueue with one buffer so the hypervisor can | |
312 | * use it to signal us later. | |
313 | */ | |
314 | sg_init_one(&sg, vb->stats, sizeof vb->stats); | |
946cfe0e | 315 | if (virtqueue_add_buf(vb->stats_vq, &sg, 1, 0, vb) < 0) |
9564e138 | 316 | BUG(); |
946cfe0e | 317 | virtqueue_kick(vb->stats_vq); |
9564e138 | 318 | } |
6b35e407 RR |
319 | |
320 | vb->thread = kthread_run(balloon, vb, "vballoon"); | |
321 | if (IS_ERR(vb->thread)) { | |
322 | err = PTR_ERR(vb->thread); | |
d2a7ddda | 323 | goto out_del_vqs; |
6b35e407 RR |
324 | } |
325 | ||
6b35e407 RR |
326 | return 0; |
327 | ||
d2a7ddda MT |
328 | out_del_vqs: |
329 | vdev->config->del_vqs(vdev); | |
6b35e407 RR |
330 | out_free_vb: |
331 | kfree(vb); | |
332 | out: | |
333 | return err; | |
334 | } | |
335 | ||
1e65175c | 336 | static void __devexit virtballoon_remove(struct virtio_device *vdev) |
6b35e407 RR |
337 | { |
338 | struct virtio_balloon *vb = vdev->priv; | |
339 | ||
340 | kthread_stop(vb->thread); | |
341 | ||
342 | /* There might be pages left in the balloon: free them. */ | |
343 | while (vb->num_pages) | |
344 | leak_balloon(vb, vb->num_pages); | |
345 | ||
346 | /* Now we reset the device so we can clean up the queues. */ | |
347 | vdev->config->reset(vdev); | |
348 | ||
d2a7ddda | 349 | vdev->config->del_vqs(vdev); |
6b35e407 RR |
350 | kfree(vb); |
351 | } | |
352 | ||
9564e138 AL |
353 | static unsigned int features[] = { |
354 | VIRTIO_BALLOON_F_MUST_TELL_HOST, | |
355 | VIRTIO_BALLOON_F_STATS_VQ, | |
356 | }; | |
c45a6816 | 357 | |
d817cd52 | 358 | static struct virtio_driver virtio_balloon_driver = { |
c45a6816 RR |
359 | .feature_table = features, |
360 | .feature_table_size = ARRAY_SIZE(features), | |
6b35e407 RR |
361 | .driver.name = KBUILD_MODNAME, |
362 | .driver.owner = THIS_MODULE, | |
363 | .id_table = id_table, | |
364 | .probe = virtballoon_probe, | |
365 | .remove = __devexit_p(virtballoon_remove), | |
366 | .config_changed = virtballoon_changed, | |
367 | }; | |
368 | ||
369 | static int __init init(void) | |
370 | { | |
d817cd52 | 371 | return register_virtio_driver(&virtio_balloon_driver); |
6b35e407 RR |
372 | } |
373 | ||
374 | static void __exit fini(void) | |
375 | { | |
d817cd52 | 376 | unregister_virtio_driver(&virtio_balloon_driver); |
6b35e407 RR |
377 | } |
378 | module_init(init); | |
379 | module_exit(fini); | |
380 | ||
381 | MODULE_DEVICE_TABLE(virtio, id_table); | |
382 | MODULE_DESCRIPTION("Virtio balloon driver"); | |
383 | MODULE_LICENSE("GPL"); |