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