Commit | Line | Data |
---|---|---|
1e214a5c SL |
1 | /* |
2 | * Virtio balloon implementation, inspired by Dor Laor and Marcelo | |
6b35e407 RR |
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 | */ | |
1e214a5c | 21 | |
6b35e407 RR |
22 | #include <linux/virtio.h> |
23 | #include <linux/virtio_balloon.h> | |
24 | #include <linux/swap.h> | |
fad7b7b2 | 25 | #include <linux/workqueue.h> |
6659a0f0 | 26 | #include <linux/delay.h> |
5a0e3ad6 | 27 | #include <linux/slab.h> |
b5a2c4f1 | 28 | #include <linux/module.h> |
e2250429 | 29 | #include <linux/balloon_compaction.h> |
5a10b7db | 30 | #include <linux/oom.h> |
3d2a3774 | 31 | #include <linux/wait.h> |
5057dcd0 | 32 | #include <linux/mm.h> |
b1123ea6 | 33 | #include <linux/mount.h> |
50d34394 | 34 | #include <linux/magic.h> |
6b35e407 | 35 | |
3ccc9372 MT |
36 | /* |
37 | * Balloon device works in 4K page units. So each page is pointed to by | |
38 | * multiple balloon pages. All memory counters in this driver are in balloon | |
39 | * page units. | |
40 | */ | |
e2250429 RA |
41 | #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT) |
42 | #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256 | |
5a10b7db RM |
43 | #define OOM_VBALLOON_DEFAULT_PAGES 256 |
44 | #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80 | |
45 | ||
46 | static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES; | |
47 | module_param(oom_pages, int, S_IRUSR | S_IWUSR); | |
48 | MODULE_PARM_DESC(oom_pages, "pages to free on OOM"); | |
3ccc9372 | 49 | |
b1123ea6 MK |
50 | #ifdef CONFIG_BALLOON_COMPACTION |
51 | static struct vfsmount *balloon_mnt; | |
52 | #endif | |
53 | ||
25e65e4e | 54 | struct virtio_balloon { |
6b35e407 | 55 | struct virtio_device *vdev; |
9564e138 | 56 | struct virtqueue *inflate_vq, *deflate_vq, *stats_vq; |
6b35e407 | 57 | |
fad7b7b2 | 58 | /* The balloon servicing is delegated to a freezable workqueue. */ |
fd0e21c3 PM |
59 | struct work_struct update_balloon_stats_work; |
60 | struct work_struct update_balloon_size_work; | |
6b35e407 | 61 | |
fad7b7b2 PM |
62 | /* Prevent updating balloon when it is being canceled. */ |
63 | spinlock_t stop_update_lock; | |
64 | bool stop_update; | |
6b35e407 RR |
65 | |
66 | /* Waiting for host to ack the pages we released. */ | |
9c378abc | 67 | wait_queue_head_t acked; |
6b35e407 | 68 | |
3ccc9372 | 69 | /* Number of balloon pages we've told the Host we're not using. */ |
6b35e407 | 70 | unsigned int num_pages; |
3ccc9372 | 71 | /* |
e2250429 RA |
72 | * The pages we've told the Host we're not using are enqueued |
73 | * at vb_dev_info->pages list. | |
3ccc9372 MT |
74 | * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE |
75 | * to num_pages above. | |
76 | */ | |
9d1ba805 | 77 | struct balloon_dev_info vb_dev_info; |
e2250429 RA |
78 | |
79 | /* Synchronize access/update to this struct virtio_balloon elements */ | |
80 | struct mutex balloon_lock; | |
6b35e407 RR |
81 | |
82 | /* The array of pfns we tell the Host about. */ | |
83 | unsigned int num_pfns; | |
87c9403b | 84 | __virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX]; |
9564e138 AL |
85 | |
86 | /* Memory statistics */ | |
87 | struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR]; | |
5a10b7db RM |
88 | |
89 | /* To register callback in oom notifier call chain */ | |
90 | struct notifier_block nb; | |
6b35e407 RR |
91 | }; |
92 | ||
93 | static struct virtio_device_id id_table[] = { | |
94 | { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID }, | |
95 | { 0 }, | |
96 | }; | |
97 | ||
1b4aa2fa HB |
98 | static u32 page_to_balloon_pfn(struct page *page) |
99 | { | |
100 | unsigned long pfn = page_to_pfn(page); | |
101 | ||
102 | BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT); | |
103 | /* Convert pfn from Linux page size to balloon page size. */ | |
3ccc9372 MT |
104 | return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE; |
105 | } | |
106 | ||
107 | static struct page *balloon_pfn_to_page(u32 pfn) | |
108 | { | |
109 | BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE); | |
110 | return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE); | |
1b4aa2fa HB |
111 | } |
112 | ||
6b35e407 RR |
113 | static void balloon_ack(struct virtqueue *vq) |
114 | { | |
9c378abc | 115 | struct virtio_balloon *vb = vq->vdev->priv; |
6b35e407 | 116 | |
9c378abc | 117 | wake_up(&vb->acked); |
6b35e407 RR |
118 | } |
119 | ||
120 | static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq) | |
121 | { | |
122 | struct scatterlist sg; | |
9c378abc | 123 | unsigned int len; |
6b35e407 RR |
124 | |
125 | sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns); | |
126 | ||
6b35e407 | 127 | /* We should always be able to add one buffer to an empty queue. */ |
4951cc90 | 128 | virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); |
946cfe0e | 129 | virtqueue_kick(vq); |
6b35e407 RR |
130 | |
131 | /* When host has read buffer, this completes via balloon_ack */ | |
9c378abc | 132 | wait_event(vb->acked, virtqueue_get_buf(vq, &len)); |
fd0e21c3 | 133 | |
6b35e407 RR |
134 | } |
135 | ||
87c9403b MT |
136 | static void set_page_pfns(struct virtio_balloon *vb, |
137 | __virtio32 pfns[], struct page *page) | |
3ccc9372 MT |
138 | { |
139 | unsigned int i; | |
140 | ||
141 | /* Set balloon pfns pointing at this page. | |
142 | * Note that the first pfn points at start of the page. */ | |
143 | for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++) | |
87c9403b MT |
144 | pfns[i] = cpu_to_virtio32(vb->vdev, |
145 | page_to_balloon_pfn(page) + i); | |
3ccc9372 MT |
146 | } |
147 | ||
fad7b7b2 | 148 | static unsigned fill_balloon(struct virtio_balloon *vb, size_t num) |
6b35e407 | 149 | { |
9d1ba805 | 150 | struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info; |
fad7b7b2 | 151 | unsigned num_allocated_pages; |
e2250429 | 152 | |
6b35e407 RR |
153 | /* We can only do one array worth at a time. */ |
154 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
155 | ||
e2250429 | 156 | mutex_lock(&vb->balloon_lock); |
3ccc9372 MT |
157 | for (vb->num_pfns = 0; vb->num_pfns < num; |
158 | vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { | |
e2250429 RA |
159 | struct page *page = balloon_page_enqueue(vb_dev_info); |
160 | ||
6b35e407 | 161 | if (!page) { |
800ba5ea | 162 | dev_info_ratelimited(&vb->vdev->dev, |
b7dfde95 LT |
163 | "Out of puff! Can't get %u pages\n", |
164 | VIRTIO_BALLOON_PAGES_PER_PAGE); | |
6b35e407 RR |
165 | /* Sleep for at least 1/5 of a second before retry. */ |
166 | msleep(200); | |
167 | break; | |
168 | } | |
87c9403b | 169 | set_page_pfns(vb, vb->pfns + vb->num_pfns, page); |
3ccc9372 | 170 | vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE; |
997e1208 DL |
171 | if (!virtio_has_feature(vb->vdev, |
172 | VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) | |
173 | adjust_managed_page_count(page, -1); | |
6b35e407 RR |
174 | } |
175 | ||
fad7b7b2 | 176 | num_allocated_pages = vb->num_pfns; |
e2250429 RA |
177 | /* Did we get any? */ |
178 | if (vb->num_pfns != 0) | |
179 | tell_host(vb, vb->inflate_vq); | |
180 | mutex_unlock(&vb->balloon_lock); | |
fad7b7b2 PM |
181 | |
182 | return num_allocated_pages; | |
6b35e407 RR |
183 | } |
184 | ||
b4d34037 | 185 | static void release_pages_balloon(struct virtio_balloon *vb) |
6b35e407 RR |
186 | { |
187 | unsigned int i; | |
87c9403b | 188 | struct page *page; |
6b35e407 | 189 | |
3ccc9372 | 190 | /* Find pfns pointing at start of each page, get pages and free them. */ |
b4d34037 | 191 | for (i = 0; i < vb->num_pfns; i += VIRTIO_BALLOON_PAGES_PER_PAGE) { |
87c9403b MT |
192 | page = balloon_pfn_to_page(virtio32_to_cpu(vb->vdev, |
193 | vb->pfns[i])); | |
997e1208 DL |
194 | if (!virtio_has_feature(vb->vdev, |
195 | VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) | |
196 | adjust_managed_page_count(page, 1); | |
d6d86c0a | 197 | put_page(page); /* balloon reference */ |
6b35e407 RR |
198 | } |
199 | } | |
200 | ||
1fd9c672 | 201 | static unsigned leak_balloon(struct virtio_balloon *vb, size_t num) |
6b35e407 | 202 | { |
1fd9c672 | 203 | unsigned num_freed_pages; |
6b35e407 | 204 | struct page *page; |
9d1ba805 | 205 | struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info; |
6b35e407 RR |
206 | |
207 | /* We can only do one array worth at a time. */ | |
208 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
209 | ||
e2250429 | 210 | mutex_lock(&vb->balloon_lock); |
37cf99e0 KN |
211 | /* We can't release more pages than taken */ |
212 | num = min(num, (size_t)vb->num_pages); | |
3ccc9372 MT |
213 | for (vb->num_pfns = 0; vb->num_pfns < num; |
214 | vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { | |
e2250429 RA |
215 | page = balloon_page_dequeue(vb_dev_info); |
216 | if (!page) | |
217 | break; | |
87c9403b | 218 | set_page_pfns(vb, vb->pfns + vb->num_pfns, page); |
3ccc9372 | 219 | vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE; |
6b35e407 RR |
220 | } |
221 | ||
1fd9c672 | 222 | num_freed_pages = vb->num_pfns; |
bf50e69f DH |
223 | /* |
224 | * Note that if | |
225 | * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST); | |
226 | * is true, we *have* to do it in this order | |
227 | */ | |
8c6bab4f LC |
228 | if (vb->num_pfns != 0) |
229 | tell_host(vb, vb->deflate_vq); | |
b4d34037 | 230 | release_pages_balloon(vb); |
f68b992b | 231 | mutex_unlock(&vb->balloon_lock); |
1fd9c672 | 232 | return num_freed_pages; |
6b35e407 RR |
233 | } |
234 | ||
9564e138 AL |
235 | static inline void update_stat(struct virtio_balloon *vb, int idx, |
236 | u16 tag, u64 val) | |
237 | { | |
238 | BUG_ON(idx >= VIRTIO_BALLOON_S_NR); | |
df81b29c MT |
239 | vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag); |
240 | vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val); | |
9564e138 AL |
241 | } |
242 | ||
243 | #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT) | |
244 | ||
245 | static void update_balloon_stats(struct virtio_balloon *vb) | |
246 | { | |
247 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
248 | struct sysinfo i; | |
249 | int idx = 0; | |
5057dcd0 | 250 | long available; |
9564e138 AL |
251 | |
252 | all_vm_events(events); | |
253 | si_meminfo(&i); | |
254 | ||
5057dcd0 IR |
255 | available = si_mem_available(); |
256 | ||
9564e138 AL |
257 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN, |
258 | pages_to_bytes(events[PSWPIN])); | |
259 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT, | |
260 | pages_to_bytes(events[PSWPOUT])); | |
261 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]); | |
262 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]); | |
263 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE, | |
264 | pages_to_bytes(i.freeram)); | |
265 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT, | |
266 | pages_to_bytes(i.totalram)); | |
5057dcd0 IR |
267 | update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL, |
268 | pages_to_bytes(available)); | |
9564e138 AL |
269 | } |
270 | ||
271 | /* | |
272 | * While most virtqueues communicate guest-initiated requests to the hypervisor, | |
273 | * the stats queue operates in reverse. The driver initializes the virtqueue | |
274 | * with a single buffer. From that point forward, all conversations consist of | |
275 | * a hypervisor request (a call to this function) which directs us to refill | |
1f34c71a | 276 | * the virtqueue with a fresh stats buffer. Since stats collection can sleep, |
fad7b7b2 PM |
277 | * we delegate the job to a freezable workqueue that will do the actual work via |
278 | * stats_handle_request(). | |
9564e138 | 279 | */ |
1f34c71a | 280 | static void stats_request(struct virtqueue *vq) |
9564e138 | 281 | { |
9c378abc | 282 | struct virtio_balloon *vb = vq->vdev->priv; |
9564e138 | 283 | |
fad7b7b2 PM |
284 | spin_lock(&vb->stop_update_lock); |
285 | if (!vb->stop_update) | |
fd0e21c3 | 286 | queue_work(system_freezable_wq, &vb->update_balloon_stats_work); |
fad7b7b2 | 287 | spin_unlock(&vb->stop_update_lock); |
1f34c71a AL |
288 | } |
289 | ||
290 | static void stats_handle_request(struct virtio_balloon *vb) | |
291 | { | |
292 | struct virtqueue *vq; | |
293 | struct scatterlist sg; | |
9c378abc | 294 | unsigned int len; |
9564e138 AL |
295 | |
296 | update_balloon_stats(vb); | |
297 | ||
1f34c71a | 298 | vq = vb->stats_vq; |
9c378abc MT |
299 | if (!virtqueue_get_buf(vq, &len)) |
300 | return; | |
9564e138 | 301 | sg_init_one(&sg, vb->stats, sizeof(vb->stats)); |
4951cc90 | 302 | virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); |
946cfe0e | 303 | virtqueue_kick(vq); |
9564e138 AL |
304 | } |
305 | ||
6b35e407 RR |
306 | static void virtballoon_changed(struct virtio_device *vdev) |
307 | { | |
308 | struct virtio_balloon *vb = vdev->priv; | |
fad7b7b2 | 309 | unsigned long flags; |
6b35e407 | 310 | |
fad7b7b2 PM |
311 | spin_lock_irqsave(&vb->stop_update_lock, flags); |
312 | if (!vb->stop_update) | |
fd0e21c3 | 313 | queue_work(system_freezable_wq, &vb->update_balloon_size_work); |
fad7b7b2 | 314 | spin_unlock_irqrestore(&vb->stop_update_lock, flags); |
6b35e407 RR |
315 | } |
316 | ||
bdc1681c | 317 | static inline s64 towards_target(struct virtio_balloon *vb) |
6b35e407 | 318 | { |
1a87228f | 319 | s64 target; |
df81b29c | 320 | u32 num_pages; |
1a87228f | 321 | |
df81b29c MT |
322 | virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages, |
323 | &num_pages); | |
855e0c52 | 324 | |
df81b29c MT |
325 | /* Legacy balloon config space is LE, unlike all other devices. */ |
326 | if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1)) | |
327 | num_pages = le32_to_cpu((__force __le32)num_pages); | |
328 | ||
329 | target = num_pages; | |
1a87228f | 330 | return target - vb->num_pages; |
6b35e407 RR |
331 | } |
332 | ||
333 | static void update_balloon_size(struct virtio_balloon *vb) | |
334 | { | |
df81b29c MT |
335 | u32 actual = vb->num_pages; |
336 | ||
337 | /* Legacy balloon config space is LE, unlike all other devices. */ | |
338 | if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1)) | |
339 | actual = (__force u32)cpu_to_le32(actual); | |
6b35e407 | 340 | |
3459f11a | 341 | virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual, |
855e0c52 | 342 | &actual); |
6b35e407 RR |
343 | } |
344 | ||
5a10b7db RM |
345 | /* |
346 | * virtballoon_oom_notify - release pages when system is under severe | |
347 | * memory pressure (called from out_of_memory()) | |
348 | * @self : notifier block struct | |
349 | * @dummy: not used | |
350 | * @parm : returned - number of freed pages | |
351 | * | |
352 | * The balancing of memory by use of the virtio balloon should not cause | |
353 | * the termination of processes while there are pages in the balloon. | |
354 | * If virtio balloon manages to release some memory, it will make the | |
355 | * system return and retry the allocation that forced the OOM killer | |
356 | * to run. | |
357 | */ | |
358 | static int virtballoon_oom_notify(struct notifier_block *self, | |
359 | unsigned long dummy, void *parm) | |
360 | { | |
361 | struct virtio_balloon *vb; | |
362 | unsigned long *freed; | |
363 | unsigned num_freed_pages; | |
364 | ||
365 | vb = container_of(self, struct virtio_balloon, nb); | |
366 | if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) | |
367 | return NOTIFY_OK; | |
368 | ||
369 | freed = parm; | |
370 | num_freed_pages = leak_balloon(vb, oom_pages); | |
371 | update_balloon_size(vb); | |
372 | *freed += num_freed_pages; | |
373 | ||
374 | return NOTIFY_OK; | |
375 | } | |
376 | ||
fd0e21c3 | 377 | static void update_balloon_stats_func(struct work_struct *work) |
6b35e407 | 378 | { |
fd0e21c3 | 379 | struct virtio_balloon *vb; |
3d2a3774 | 380 | |
fd0e21c3 PM |
381 | vb = container_of(work, struct virtio_balloon, |
382 | update_balloon_stats_work); | |
383 | stats_handle_request(vb); | |
384 | } | |
1f74ef0f | 385 | |
fd0e21c3 | 386 | static void update_balloon_size_func(struct work_struct *work) |
6b35e407 | 387 | { |
fad7b7b2 PM |
388 | struct virtio_balloon *vb; |
389 | s64 diff; | |
3d2a3774 | 390 | |
fd0e21c3 PM |
391 | vb = container_of(work, struct virtio_balloon, |
392 | update_balloon_size_work); | |
fad7b7b2 | 393 | diff = towards_target(vb); |
1f74ef0f | 394 | |
fad7b7b2 PM |
395 | if (diff > 0) |
396 | diff -= fill_balloon(vb, diff); | |
397 | else if (diff < 0) | |
398 | diff += leak_balloon(vb, -diff); | |
399 | update_balloon_size(vb); | |
400 | ||
401 | if (diff) | |
402 | queue_work(system_freezable_wq, work); | |
6b35e407 RR |
403 | } |
404 | ||
be91c33d | 405 | static int init_vqs(struct virtio_balloon *vb) |
6b35e407 | 406 | { |
9564e138 | 407 | struct virtqueue *vqs[3]; |
1f34c71a | 408 | vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request }; |
f7ad26ff | 409 | static const char * const names[] = { "inflate", "deflate", "stats" }; |
9564e138 | 410 | int err, nvqs; |
6b35e407 | 411 | |
be91c33d AS |
412 | /* |
413 | * We expect two virtqueues: inflate and deflate, and | |
414 | * optionally stat. | |
415 | */ | |
9564e138 | 416 | nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2; |
fb5e31d9 CH |
417 | err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names, |
418 | NULL); | |
d2a7ddda | 419 | if (err) |
be91c33d | 420 | return err; |
6b35e407 | 421 | |
d2a7ddda MT |
422 | vb->inflate_vq = vqs[0]; |
423 | vb->deflate_vq = vqs[1]; | |
9564e138 AL |
424 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { |
425 | struct scatterlist sg; | |
426 | vb->stats_vq = vqs[2]; | |
427 | ||
428 | /* | |
429 | * Prime this virtqueue with one buffer so the hypervisor can | |
4951cc90 | 430 | * use it to signal us later (it can't be broken yet!). |
9564e138 AL |
431 | */ |
432 | sg_init_one(&sg, vb->stats, sizeof vb->stats); | |
92549abc | 433 | if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL) |
f96fde41 | 434 | < 0) |
9564e138 | 435 | BUG(); |
946cfe0e | 436 | virtqueue_kick(vb->stats_vq); |
9564e138 | 437 | } |
be91c33d AS |
438 | return 0; |
439 | } | |
440 | ||
e2250429 RA |
441 | #ifdef CONFIG_BALLOON_COMPACTION |
442 | /* | |
443 | * virtballoon_migratepage - perform the balloon page migration on behalf of | |
444 | * a compation thread. (called under page lock) | |
9d1ba805 | 445 | * @vb_dev_info: the balloon device |
e2250429 RA |
446 | * @newpage: page that will replace the isolated page after migration finishes. |
447 | * @page : the isolated (old) page that is about to be migrated to newpage. | |
448 | * @mode : compaction mode -- not used for balloon page migration. | |
449 | * | |
450 | * After a ballooned page gets isolated by compaction procedures, this is the | |
451 | * function that performs the page migration on behalf of a compaction thread | |
452 | * The page migration for virtio balloon is done in a simple swap fashion which | |
453 | * follows these two macro steps: | |
454 | * 1) insert newpage into vb->pages list and update the host about it; | |
455 | * 2) update the host about the old page removed from vb->pages list; | |
456 | * | |
457 | * This function preforms the balloon page migration task. | |
458 | * Called through balloon_mapping->a_ops->migratepage | |
459 | */ | |
9d1ba805 | 460 | static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info, |
e2250429 RA |
461 | struct page *newpage, struct page *page, enum migrate_mode mode) |
462 | { | |
9d1ba805 KK |
463 | struct virtio_balloon *vb = container_of(vb_dev_info, |
464 | struct virtio_balloon, vb_dev_info); | |
e2250429 RA |
465 | unsigned long flags; |
466 | ||
e2250429 RA |
467 | /* |
468 | * In order to avoid lock contention while migrating pages concurrently | |
469 | * to leak_balloon() or fill_balloon() we just give up the balloon_lock | |
470 | * this turn, as it is easier to retry the page migration later. | |
471 | * This also prevents fill_balloon() getting stuck into a mutex | |
472 | * recursion in the case it ends up triggering memory compaction | |
473 | * while it is attempting to inflate the ballon. | |
474 | */ | |
475 | if (!mutex_trylock(&vb->balloon_lock)) | |
476 | return -EAGAIN; | |
477 | ||
d6d86c0a KK |
478 | get_page(newpage); /* balloon reference */ |
479 | ||
e2250429 RA |
480 | /* balloon's page migration 1st step -- inflate "newpage" */ |
481 | spin_lock_irqsave(&vb_dev_info->pages_lock, flags); | |
9d1ba805 | 482 | balloon_page_insert(vb_dev_info, newpage); |
e2250429 | 483 | vb_dev_info->isolated_pages--; |
09316c09 | 484 | __count_vm_event(BALLOON_MIGRATE); |
e2250429 RA |
485 | spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags); |
486 | vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; | |
87c9403b | 487 | set_page_pfns(vb, vb->pfns, newpage); |
e2250429 RA |
488 | tell_host(vb, vb->inflate_vq); |
489 | ||
d6d86c0a | 490 | /* balloon's page migration 2nd step -- deflate "page" */ |
e2250429 RA |
491 | balloon_page_delete(page); |
492 | vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; | |
87c9403b | 493 | set_page_pfns(vb, vb->pfns, page); |
e2250429 RA |
494 | tell_host(vb, vb->deflate_vq); |
495 | ||
496 | mutex_unlock(&vb->balloon_lock); | |
497 | ||
d6d86c0a KK |
498 | put_page(page); /* balloon reference */ |
499 | ||
dd4123f3 | 500 | return MIGRATEPAGE_SUCCESS; |
e2250429 | 501 | } |
b1123ea6 MK |
502 | |
503 | static struct dentry *balloon_mount(struct file_system_type *fs_type, | |
504 | int flags, const char *dev_name, void *data) | |
505 | { | |
506 | static const struct dentry_operations ops = { | |
507 | .d_dname = simple_dname, | |
508 | }; | |
509 | ||
510 | return mount_pseudo(fs_type, "balloon-kvm:", NULL, &ops, | |
511 | BALLOON_KVM_MAGIC); | |
512 | } | |
513 | ||
514 | static struct file_system_type balloon_fs = { | |
515 | .name = "balloon-kvm", | |
516 | .mount = balloon_mount, | |
517 | .kill_sb = kill_anon_super, | |
518 | }; | |
519 | ||
e2250429 RA |
520 | #endif /* CONFIG_BALLOON_COMPACTION */ |
521 | ||
be91c33d AS |
522 | static int virtballoon_probe(struct virtio_device *vdev) |
523 | { | |
524 | struct virtio_balloon *vb; | |
525 | int err; | |
526 | ||
2d9becc1 MT |
527 | if (!vdev->config->get) { |
528 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
529 | __func__); | |
530 | return -EINVAL; | |
531 | } | |
532 | ||
be91c33d AS |
533 | vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL); |
534 | if (!vb) { | |
535 | err = -ENOMEM; | |
536 | goto out; | |
537 | } | |
538 | ||
fd0e21c3 PM |
539 | INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func); |
540 | INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func); | |
fad7b7b2 PM |
541 | spin_lock_init(&vb->stop_update_lock); |
542 | vb->stop_update = false; | |
be91c33d | 543 | vb->num_pages = 0; |
e2250429 | 544 | mutex_init(&vb->balloon_lock); |
9c378abc | 545 | init_waitqueue_head(&vb->acked); |
be91c33d | 546 | vb->vdev = vdev; |
be91c33d | 547 | |
9d1ba805 | 548 | balloon_devinfo_init(&vb->vb_dev_info); |
e2250429 | 549 | |
be91c33d AS |
550 | err = init_vqs(vb); |
551 | if (err) | |
9d1ba805 | 552 | goto out_free_vb; |
6b35e407 | 553 | |
5a10b7db RM |
554 | vb->nb.notifier_call = virtballoon_oom_notify; |
555 | vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY; | |
556 | err = register_oom_notifier(&vb->nb); | |
557 | if (err < 0) | |
b1123ea6 MK |
558 | goto out_del_vqs; |
559 | ||
560 | #ifdef CONFIG_BALLOON_COMPACTION | |
561 | balloon_mnt = kern_mount(&balloon_fs); | |
562 | if (IS_ERR(balloon_mnt)) { | |
563 | err = PTR_ERR(balloon_mnt); | |
564 | unregister_oom_notifier(&vb->nb); | |
565 | goto out_del_vqs; | |
566 | } | |
567 | ||
568 | vb->vb_dev_info.migratepage = virtballoon_migratepage; | |
569 | vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb); | |
570 | if (IS_ERR(vb->vb_dev_info.inode)) { | |
571 | err = PTR_ERR(vb->vb_dev_info.inode); | |
572 | kern_unmount(balloon_mnt); | |
573 | unregister_oom_notifier(&vb->nb); | |
574 | vb->vb_dev_info.inode = NULL; | |
575 | goto out_del_vqs; | |
576 | } | |
577 | vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops; | |
578 | #endif | |
5a10b7db | 579 | |
88660f7f MT |
580 | virtio_device_ready(vdev); |
581 | ||
8424af53 KN |
582 | if (towards_target(vb)) |
583 | virtballoon_changed(vdev); | |
6b35e407 RR |
584 | return 0; |
585 | ||
b1123ea6 | 586 | out_del_vqs: |
d2a7ddda | 587 | vdev->config->del_vqs(vdev); |
6b35e407 RR |
588 | out_free_vb: |
589 | kfree(vb); | |
590 | out: | |
591 | return err; | |
592 | } | |
593 | ||
c877bab5 | 594 | static void remove_common(struct virtio_balloon *vb) |
6b35e407 | 595 | { |
6b35e407 RR |
596 | /* There might be pages left in the balloon: free them. */ |
597 | while (vb->num_pages) | |
598 | leak_balloon(vb, vb->num_pages); | |
b8ae0eb3 | 599 | update_balloon_size(vb); |
6b35e407 RR |
600 | |
601 | /* Now we reset the device so we can clean up the queues. */ | |
c877bab5 | 602 | vb->vdev->config->reset(vb->vdev); |
6b35e407 | 603 | |
c877bab5 AS |
604 | vb->vdev->config->del_vqs(vb->vdev); |
605 | } | |
606 | ||
8590dbc7 | 607 | static void virtballoon_remove(struct virtio_device *vdev) |
c877bab5 AS |
608 | { |
609 | struct virtio_balloon *vb = vdev->priv; | |
610 | ||
5a10b7db | 611 | unregister_oom_notifier(&vb->nb); |
fad7b7b2 PM |
612 | |
613 | spin_lock_irq(&vb->stop_update_lock); | |
614 | vb->stop_update = true; | |
615 | spin_unlock_irq(&vb->stop_update_lock); | |
fd0e21c3 PM |
616 | cancel_work_sync(&vb->update_balloon_size_work); |
617 | cancel_work_sync(&vb->update_balloon_stats_work); | |
fad7b7b2 | 618 | |
c877bab5 | 619 | remove_common(vb); |
9c57b580 | 620 | #ifdef CONFIG_BALLOON_COMPACTION |
b1123ea6 MK |
621 | if (vb->vb_dev_info.inode) |
622 | iput(vb->vb_dev_info.inode); | |
9c57b580 YX |
623 | |
624 | kern_unmount(balloon_mnt); | |
625 | #endif | |
6b35e407 RR |
626 | kfree(vb); |
627 | } | |
628 | ||
89107000 | 629 | #ifdef CONFIG_PM_SLEEP |
e562966d AS |
630 | static int virtballoon_freeze(struct virtio_device *vdev) |
631 | { | |
4eb05d56 AS |
632 | struct virtio_balloon *vb = vdev->priv; |
633 | ||
e562966d | 634 | /* |
fad7b7b2 | 635 | * The workqueue is already frozen by the PM core before this |
e562966d AS |
636 | * function is called. |
637 | */ | |
c877bab5 | 638 | remove_common(vb); |
e562966d AS |
639 | return 0; |
640 | } | |
641 | ||
c45b4166 | 642 | static int virtballoon_restore(struct virtio_device *vdev) |
4eb05d56 AS |
643 | { |
644 | struct virtio_balloon *vb = vdev->priv; | |
645 | int ret; | |
646 | ||
647 | ret = init_vqs(vdev->priv); | |
648 | if (ret) | |
649 | return ret; | |
650 | ||
486d2e63 MT |
651 | virtio_device_ready(vdev); |
652 | ||
fad7b7b2 PM |
653 | if (towards_target(vb)) |
654 | virtballoon_changed(vdev); | |
4eb05d56 AS |
655 | update_balloon_size(vb); |
656 | return 0; | |
657 | } | |
e562966d AS |
658 | #endif |
659 | ||
9564e138 AL |
660 | static unsigned int features[] = { |
661 | VIRTIO_BALLOON_F_MUST_TELL_HOST, | |
662 | VIRTIO_BALLOON_F_STATS_VQ, | |
5a10b7db | 663 | VIRTIO_BALLOON_F_DEFLATE_ON_OOM, |
9564e138 | 664 | }; |
c45a6816 | 665 | |
d817cd52 | 666 | static struct virtio_driver virtio_balloon_driver = { |
c45a6816 RR |
667 | .feature_table = features, |
668 | .feature_table_size = ARRAY_SIZE(features), | |
6b35e407 RR |
669 | .driver.name = KBUILD_MODNAME, |
670 | .driver.owner = THIS_MODULE, | |
671 | .id_table = id_table, | |
672 | .probe = virtballoon_probe, | |
8590dbc7 | 673 | .remove = virtballoon_remove, |
6b35e407 | 674 | .config_changed = virtballoon_changed, |
89107000 | 675 | #ifdef CONFIG_PM_SLEEP |
e562966d AS |
676 | .freeze = virtballoon_freeze, |
677 | .restore = virtballoon_restore, | |
e562966d | 678 | #endif |
6b35e407 RR |
679 | }; |
680 | ||
b2a17029 | 681 | module_virtio_driver(virtio_balloon_driver); |
6b35e407 RR |
682 | MODULE_DEVICE_TABLE(virtio, id_table); |
683 | MODULE_DESCRIPTION("Virtio balloon driver"); | |
684 | MODULE_LICENSE("GPL"); |