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 | ||
43 | /* Do we have to tell Host *before* we reuse pages? */ | |
44 | bool tell_host_first; | |
45 | ||
46 | /* The pages we've told the Host we're not using. */ | |
47 | unsigned int num_pages; | |
48 | struct list_head pages; | |
49 | ||
50 | /* The array of pfns we tell the Host about. */ | |
51 | unsigned int num_pfns; | |
52 | u32 pfns[256]; | |
9564e138 AL |
53 | |
54 | /* Memory statistics */ | |
1f34c71a | 55 | int need_stats_update; |
9564e138 | 56 | struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR]; |
6b35e407 RR |
57 | }; |
58 | ||
59 | static struct virtio_device_id id_table[] = { | |
60 | { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID }, | |
61 | { 0 }, | |
62 | }; | |
63 | ||
1b4aa2fa HB |
64 | static u32 page_to_balloon_pfn(struct page *page) |
65 | { | |
66 | unsigned long pfn = page_to_pfn(page); | |
67 | ||
68 | BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT); | |
69 | /* Convert pfn from Linux page size to balloon page size. */ | |
70 | return pfn >> (PAGE_SHIFT - VIRTIO_BALLOON_PFN_SHIFT); | |
71 | } | |
72 | ||
6b35e407 RR |
73 | static void balloon_ack(struct virtqueue *vq) |
74 | { | |
75 | struct virtio_balloon *vb; | |
76 | unsigned int len; | |
77 | ||
946cfe0e | 78 | vb = virtqueue_get_buf(vq, &len); |
6b35e407 RR |
79 | if (vb) |
80 | complete(&vb->acked); | |
81 | } | |
82 | ||
83 | static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq) | |
84 | { | |
85 | struct scatterlist sg; | |
86 | ||
87 | sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns); | |
88 | ||
89 | init_completion(&vb->acked); | |
90 | ||
91 | /* We should always be able to add one buffer to an empty queue. */ | |
946cfe0e | 92 | if (virtqueue_add_buf(vq, &sg, 1, 0, vb) < 0) |
6b35e407 | 93 | BUG(); |
946cfe0e | 94 | virtqueue_kick(vq); |
6b35e407 RR |
95 | |
96 | /* When host has read buffer, this completes via balloon_ack */ | |
97 | wait_for_completion(&vb->acked); | |
98 | } | |
99 | ||
100 | static void fill_balloon(struct virtio_balloon *vb, size_t num) | |
101 | { | |
102 | /* We can only do one array worth at a time. */ | |
103 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
104 | ||
105 | for (vb->num_pfns = 0; vb->num_pfns < num; vb->num_pfns++) { | |
61fb06cc BS |
106 | struct page *page = alloc_page(GFP_HIGHUSER | __GFP_NORETRY | |
107 | __GFP_NOMEMALLOC | __GFP_NOWARN); | |
6b35e407 RR |
108 | if (!page) { |
109 | if (printk_ratelimit()) | |
110 | dev_printk(KERN_INFO, &vb->vdev->dev, | |
111 | "Out of puff! Can't get %zu pages\n", | |
112 | num); | |
113 | /* Sleep for at least 1/5 of a second before retry. */ | |
114 | msleep(200); | |
115 | break; | |
116 | } | |
1b4aa2fa | 117 | vb->pfns[vb->num_pfns] = page_to_balloon_pfn(page); |
6b35e407 RR |
118 | totalram_pages--; |
119 | vb->num_pages++; | |
120 | list_add(&page->lru, &vb->pages); | |
121 | } | |
122 | ||
123 | /* Didn't get any? Oh well. */ | |
124 | if (vb->num_pfns == 0) | |
125 | return; | |
126 | ||
127 | tell_host(vb, vb->inflate_vq); | |
128 | } | |
129 | ||
130 | static void release_pages_by_pfn(const u32 pfns[], unsigned int num) | |
131 | { | |
132 | unsigned int i; | |
133 | ||
134 | for (i = 0; i < num; i++) { | |
135 | __free_page(pfn_to_page(pfns[i])); | |
136 | totalram_pages++; | |
137 | } | |
138 | } | |
139 | ||
140 | static void leak_balloon(struct virtio_balloon *vb, size_t num) | |
141 | { | |
142 | struct page *page; | |
143 | ||
144 | /* We can only do one array worth at a time. */ | |
145 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
146 | ||
147 | for (vb->num_pfns = 0; vb->num_pfns < num; vb->num_pfns++) { | |
148 | page = list_first_entry(&vb->pages, struct page, lru); | |
149 | list_del(&page->lru); | |
1b4aa2fa | 150 | vb->pfns[vb->num_pfns] = page_to_balloon_pfn(page); |
6b35e407 RR |
151 | vb->num_pages--; |
152 | } | |
153 | ||
154 | if (vb->tell_host_first) { | |
155 | tell_host(vb, vb->deflate_vq); | |
156 | release_pages_by_pfn(vb->pfns, vb->num_pfns); | |
157 | } else { | |
158 | release_pages_by_pfn(vb->pfns, vb->num_pfns); | |
159 | tell_host(vb, vb->deflate_vq); | |
160 | } | |
161 | } | |
162 | ||
9564e138 AL |
163 | static inline void update_stat(struct virtio_balloon *vb, int idx, |
164 | u16 tag, u64 val) | |
165 | { | |
166 | BUG_ON(idx >= VIRTIO_BALLOON_S_NR); | |
167 | vb->stats[idx].tag = tag; | |
168 | vb->stats[idx].val = val; | |
169 | } | |
170 | ||
171 | #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT) | |
172 | ||
173 | static void update_balloon_stats(struct virtio_balloon *vb) | |
174 | { | |
175 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
176 | struct sysinfo i; | |
177 | int idx = 0; | |
178 | ||
179 | all_vm_events(events); | |
180 | si_meminfo(&i); | |
181 | ||
182 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN, | |
183 | pages_to_bytes(events[PSWPIN])); | |
184 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT, | |
185 | pages_to_bytes(events[PSWPOUT])); | |
186 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]); | |
187 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]); | |
188 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE, | |
189 | pages_to_bytes(i.freeram)); | |
190 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT, | |
191 | pages_to_bytes(i.totalram)); | |
192 | } | |
193 | ||
194 | /* | |
195 | * While most virtqueues communicate guest-initiated requests to the hypervisor, | |
196 | * the stats queue operates in reverse. The driver initializes the virtqueue | |
197 | * with a single buffer. From that point forward, all conversations consist of | |
198 | * a hypervisor request (a call to this function) which directs us to refill | |
1f34c71a AL |
199 | * the virtqueue with a fresh stats buffer. Since stats collection can sleep, |
200 | * we notify our kthread which does the actual work via stats_handle_request(). | |
9564e138 | 201 | */ |
1f34c71a | 202 | static void stats_request(struct virtqueue *vq) |
9564e138 AL |
203 | { |
204 | struct virtio_balloon *vb; | |
205 | unsigned int len; | |
9564e138 | 206 | |
946cfe0e | 207 | vb = virtqueue_get_buf(vq, &len); |
9564e138 AL |
208 | if (!vb) |
209 | return; | |
1f34c71a AL |
210 | vb->need_stats_update = 1; |
211 | wake_up(&vb->config_change); | |
212 | } | |
213 | ||
214 | static void stats_handle_request(struct virtio_balloon *vb) | |
215 | { | |
216 | struct virtqueue *vq; | |
217 | struct scatterlist sg; | |
9564e138 | 218 | |
1f34c71a | 219 | vb->need_stats_update = 0; |
9564e138 AL |
220 | update_balloon_stats(vb); |
221 | ||
1f34c71a | 222 | vq = vb->stats_vq; |
9564e138 | 223 | sg_init_one(&sg, vb->stats, sizeof(vb->stats)); |
946cfe0e | 224 | if (virtqueue_add_buf(vq, &sg, 1, 0, vb) < 0) |
9564e138 | 225 | BUG(); |
946cfe0e | 226 | virtqueue_kick(vq); |
9564e138 AL |
227 | } |
228 | ||
6b35e407 RR |
229 | static void virtballoon_changed(struct virtio_device *vdev) |
230 | { | |
231 | struct virtio_balloon *vb = vdev->priv; | |
232 | ||
233 | wake_up(&vb->config_change); | |
234 | } | |
235 | ||
bdc1681c | 236 | static inline s64 towards_target(struct virtio_balloon *vb) |
6b35e407 RR |
237 | { |
238 | u32 v; | |
72e61eb4 RR |
239 | vb->vdev->config->get(vb->vdev, |
240 | offsetof(struct virtio_balloon_config, num_pages), | |
241 | &v, sizeof(v)); | |
532a6086 | 242 | return (s64)v - vb->num_pages; |
6b35e407 RR |
243 | } |
244 | ||
245 | static void update_balloon_size(struct virtio_balloon *vb) | |
246 | { | |
247 | __le32 actual = cpu_to_le32(vb->num_pages); | |
248 | ||
249 | vb->vdev->config->set(vb->vdev, | |
250 | offsetof(struct virtio_balloon_config, actual), | |
251 | &actual, sizeof(actual)); | |
252 | } | |
253 | ||
254 | static int balloon(void *_vballoon) | |
255 | { | |
256 | struct virtio_balloon *vb = _vballoon; | |
257 | ||
258 | set_freezable(); | |
259 | while (!kthread_should_stop()) { | |
bdc1681c | 260 | s64 diff; |
6b35e407 RR |
261 | |
262 | try_to_freeze(); | |
263 | wait_event_interruptible(vb->config_change, | |
264 | (diff = towards_target(vb)) != 0 | |
1f34c71a | 265 | || vb->need_stats_update |
84a139a9 MT |
266 | || kthread_should_stop() |
267 | || freezing(current)); | |
1f34c71a AL |
268 | if (vb->need_stats_update) |
269 | stats_handle_request(vb); | |
6b35e407 RR |
270 | if (diff > 0) |
271 | fill_balloon(vb, diff); | |
272 | else if (diff < 0) | |
273 | leak_balloon(vb, -diff); | |
274 | update_balloon_size(vb); | |
275 | } | |
276 | return 0; | |
277 | } | |
278 | ||
279 | static int virtballoon_probe(struct virtio_device *vdev) | |
280 | { | |
281 | struct virtio_balloon *vb; | |
9564e138 | 282 | struct virtqueue *vqs[3]; |
1f34c71a | 283 | vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request }; |
9564e138 AL |
284 | const char *names[] = { "inflate", "deflate", "stats" }; |
285 | int err, nvqs; | |
6b35e407 RR |
286 | |
287 | vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL); | |
288 | if (!vb) { | |
289 | err = -ENOMEM; | |
290 | goto out; | |
291 | } | |
292 | ||
293 | INIT_LIST_HEAD(&vb->pages); | |
294 | vb->num_pages = 0; | |
295 | init_waitqueue_head(&vb->config_change); | |
296 | vb->vdev = vdev; | |
169c246a | 297 | vb->need_stats_update = 0; |
6b35e407 | 298 | |
9564e138 AL |
299 | /* We expect two virtqueues: inflate and deflate, |
300 | * and optionally stat. */ | |
301 | nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2; | |
302 | err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names); | |
d2a7ddda | 303 | if (err) |
6b35e407 | 304 | goto out_free_vb; |
6b35e407 | 305 | |
d2a7ddda MT |
306 | vb->inflate_vq = vqs[0]; |
307 | vb->deflate_vq = vqs[1]; | |
9564e138 AL |
308 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { |
309 | struct scatterlist sg; | |
310 | vb->stats_vq = vqs[2]; | |
311 | ||
312 | /* | |
313 | * Prime this virtqueue with one buffer so the hypervisor can | |
314 | * use it to signal us later. | |
315 | */ | |
316 | sg_init_one(&sg, vb->stats, sizeof vb->stats); | |
946cfe0e | 317 | if (virtqueue_add_buf(vb->stats_vq, &sg, 1, 0, vb) < 0) |
9564e138 | 318 | BUG(); |
946cfe0e | 319 | virtqueue_kick(vb->stats_vq); |
9564e138 | 320 | } |
6b35e407 RR |
321 | |
322 | vb->thread = kthread_run(balloon, vb, "vballoon"); | |
323 | if (IS_ERR(vb->thread)) { | |
324 | err = PTR_ERR(vb->thread); | |
d2a7ddda | 325 | goto out_del_vqs; |
6b35e407 RR |
326 | } |
327 | ||
328 | vb->tell_host_first | |
c45a6816 | 329 | = virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST); |
6b35e407 RR |
330 | |
331 | return 0; | |
332 | ||
d2a7ddda MT |
333 | out_del_vqs: |
334 | vdev->config->del_vqs(vdev); | |
6b35e407 RR |
335 | out_free_vb: |
336 | kfree(vb); | |
337 | out: | |
338 | return err; | |
339 | } | |
340 | ||
1e65175c | 341 | static void __devexit virtballoon_remove(struct virtio_device *vdev) |
6b35e407 RR |
342 | { |
343 | struct virtio_balloon *vb = vdev->priv; | |
344 | ||
345 | kthread_stop(vb->thread); | |
346 | ||
347 | /* There might be pages left in the balloon: free them. */ | |
348 | while (vb->num_pages) | |
349 | leak_balloon(vb, vb->num_pages); | |
350 | ||
351 | /* Now we reset the device so we can clean up the queues. */ | |
352 | vdev->config->reset(vdev); | |
353 | ||
d2a7ddda | 354 | vdev->config->del_vqs(vdev); |
6b35e407 RR |
355 | kfree(vb); |
356 | } | |
357 | ||
9564e138 AL |
358 | static unsigned int features[] = { |
359 | VIRTIO_BALLOON_F_MUST_TELL_HOST, | |
360 | VIRTIO_BALLOON_F_STATS_VQ, | |
361 | }; | |
c45a6816 | 362 | |
d817cd52 | 363 | static struct virtio_driver virtio_balloon_driver = { |
c45a6816 RR |
364 | .feature_table = features, |
365 | .feature_table_size = ARRAY_SIZE(features), | |
6b35e407 RR |
366 | .driver.name = KBUILD_MODNAME, |
367 | .driver.owner = THIS_MODULE, | |
368 | .id_table = id_table, | |
369 | .probe = virtballoon_probe, | |
370 | .remove = __devexit_p(virtballoon_remove), | |
371 | .config_changed = virtballoon_changed, | |
372 | }; | |
373 | ||
374 | static int __init init(void) | |
375 | { | |
d817cd52 | 376 | return register_virtio_driver(&virtio_balloon_driver); |
6b35e407 RR |
377 | } |
378 | ||
379 | static void __exit fini(void) | |
380 | { | |
d817cd52 | 381 | unregister_virtio_driver(&virtio_balloon_driver); |
6b35e407 RR |
382 | } |
383 | module_init(init); | |
384 | module_exit(fini); | |
385 | ||
386 | MODULE_DEVICE_TABLE(virtio, id_table); | |
387 | MODULE_DESCRIPTION("Virtio balloon driver"); | |
388 | MODULE_LICENSE("GPL"); |