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