]>
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> |
3d2a3774 | 30 | #include <linux/wait.h> |
5057dcd0 | 31 | #include <linux/mm.h> |
b1123ea6 | 32 | #include <linux/mount.h> |
50d34394 | 33 | #include <linux/magic.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 VIRTBALLOON_OOM_NOTIFY_PRIORITY 80 |
43 | ||
86a55978 WW |
44 | #define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \ |
45 | __GFP_NOMEMALLOC) | |
46 | /* The order of free page blocks to report to host */ | |
47 | #define VIRTIO_BALLOON_FREE_PAGE_ORDER (MAX_ORDER - 1) | |
48 | /* The size of a free page block in bytes */ | |
49 | #define VIRTIO_BALLOON_FREE_PAGE_SIZE \ | |
50 | (1 << (VIRTIO_BALLOON_FREE_PAGE_ORDER + PAGE_SHIFT)) | |
51 | ||
b1123ea6 MK |
52 | #ifdef CONFIG_BALLOON_COMPACTION |
53 | static struct vfsmount *balloon_mnt; | |
54 | #endif | |
55 | ||
86a55978 WW |
56 | enum virtio_balloon_vq { |
57 | VIRTIO_BALLOON_VQ_INFLATE, | |
58 | VIRTIO_BALLOON_VQ_DEFLATE, | |
59 | VIRTIO_BALLOON_VQ_STATS, | |
60 | VIRTIO_BALLOON_VQ_FREE_PAGE, | |
61 | VIRTIO_BALLOON_VQ_MAX | |
62 | }; | |
63 | ||
bf4dc0b2 WW |
64 | enum virtio_balloon_config_read { |
65 | VIRTIO_BALLOON_CONFIG_READ_CMD_ID = 0, | |
66 | }; | |
67 | ||
25e65e4e | 68 | struct virtio_balloon { |
6b35e407 | 69 | struct virtio_device *vdev; |
86a55978 WW |
70 | struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq; |
71 | ||
72 | /* Balloon's own wq for cpu-intensive work items */ | |
73 | struct workqueue_struct *balloon_wq; | |
74 | /* The free page reporting work item submitted to the balloon wq */ | |
75 | struct work_struct report_free_page_work; | |
6b35e407 | 76 | |
fad7b7b2 | 77 | /* The balloon servicing is delegated to a freezable workqueue. */ |
fd0e21c3 PM |
78 | struct work_struct update_balloon_stats_work; |
79 | struct work_struct update_balloon_size_work; | |
6b35e407 | 80 | |
fad7b7b2 PM |
81 | /* Prevent updating balloon when it is being canceled. */ |
82 | spinlock_t stop_update_lock; | |
83 | bool stop_update; | |
bf4dc0b2 WW |
84 | /* Bitmap to indicate if reading the related config fields are needed */ |
85 | unsigned long config_read_bitmap; | |
6b35e407 | 86 | |
86a55978 WW |
87 | /* The list of allocated free pages, waiting to be given back to mm */ |
88 | struct list_head free_page_list; | |
89 | spinlock_t free_page_list_lock; | |
90 | /* The number of free page blocks on the above list */ | |
91 | unsigned long num_free_page_blocks; | |
bf4dc0b2 WW |
92 | /* |
93 | * The cmd id received from host. | |
94 | * Read it via virtio_balloon_cmd_id_received to get the latest value | |
95 | * sent from host. | |
96 | */ | |
97 | u32 cmd_id_received_cache; | |
86a55978 WW |
98 | /* The cmd id that is actively in use */ |
99 | __virtio32 cmd_id_active; | |
100 | /* Buffer to store the stop sign */ | |
101 | __virtio32 cmd_id_stop; | |
102 | ||
6b35e407 | 103 | /* Waiting for host to ack the pages we released. */ |
9c378abc | 104 | wait_queue_head_t acked; |
6b35e407 | 105 | |
3ccc9372 | 106 | /* Number of balloon pages we've told the Host we're not using. */ |
6b35e407 | 107 | unsigned int num_pages; |
3ccc9372 | 108 | /* |
e2250429 RA |
109 | * The pages we've told the Host we're not using are enqueued |
110 | * at vb_dev_info->pages list. | |
3ccc9372 MT |
111 | * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE |
112 | * to num_pages above. | |
113 | */ | |
9d1ba805 | 114 | struct balloon_dev_info vb_dev_info; |
e2250429 RA |
115 | |
116 | /* Synchronize access/update to this struct virtio_balloon elements */ | |
117 | struct mutex balloon_lock; | |
6b35e407 RR |
118 | |
119 | /* The array of pfns we tell the Host about. */ | |
120 | unsigned int num_pfns; | |
87c9403b | 121 | __virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX]; |
9564e138 AL |
122 | |
123 | /* Memory statistics */ | |
124 | struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR]; | |
5a10b7db | 125 | |
71994620 WW |
126 | /* To register a shrinker to shrink memory upon memory pressure */ |
127 | struct shrinker shrinker; | |
6b35e407 RR |
128 | }; |
129 | ||
130 | static struct virtio_device_id id_table[] = { | |
131 | { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID }, | |
132 | { 0 }, | |
133 | }; | |
134 | ||
1b4aa2fa HB |
135 | static u32 page_to_balloon_pfn(struct page *page) |
136 | { | |
137 | unsigned long pfn = page_to_pfn(page); | |
138 | ||
139 | BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT); | |
140 | /* Convert pfn from Linux page size to balloon page size. */ | |
3ccc9372 MT |
141 | return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE; |
142 | } | |
143 | ||
6b35e407 RR |
144 | static void balloon_ack(struct virtqueue *vq) |
145 | { | |
9c378abc | 146 | struct virtio_balloon *vb = vq->vdev->priv; |
6b35e407 | 147 | |
9c378abc | 148 | wake_up(&vb->acked); |
6b35e407 RR |
149 | } |
150 | ||
151 | static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq) | |
152 | { | |
153 | struct scatterlist sg; | |
9c378abc | 154 | unsigned int len; |
6b35e407 RR |
155 | |
156 | sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns); | |
157 | ||
6b35e407 | 158 | /* We should always be able to add one buffer to an empty queue. */ |
4951cc90 | 159 | virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); |
946cfe0e | 160 | virtqueue_kick(vq); |
6b35e407 RR |
161 | |
162 | /* When host has read buffer, this completes via balloon_ack */ | |
9c378abc | 163 | wait_event(vb->acked, virtqueue_get_buf(vq, &len)); |
fd0e21c3 | 164 | |
6b35e407 RR |
165 | } |
166 | ||
87c9403b MT |
167 | static void set_page_pfns(struct virtio_balloon *vb, |
168 | __virtio32 pfns[], struct page *page) | |
3ccc9372 MT |
169 | { |
170 | unsigned int i; | |
171 | ||
f9aada5f WW |
172 | /* |
173 | * Set balloon pfns pointing at this page. | |
174 | * Note that the first pfn points at start of the page. | |
175 | */ | |
3ccc9372 | 176 | for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++) |
87c9403b MT |
177 | pfns[i] = cpu_to_virtio32(vb->vdev, |
178 | page_to_balloon_pfn(page) + i); | |
3ccc9372 MT |
179 | } |
180 | ||
fad7b7b2 | 181 | static unsigned fill_balloon(struct virtio_balloon *vb, size_t num) |
6b35e407 | 182 | { |
fad7b7b2 | 183 | unsigned num_allocated_pages; |
c7cdff0e MT |
184 | unsigned num_pfns; |
185 | struct page *page; | |
186 | LIST_HEAD(pages); | |
e2250429 | 187 | |
6b35e407 RR |
188 | /* We can only do one array worth at a time. */ |
189 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
190 | ||
c7cdff0e MT |
191 | for (num_pfns = 0; num_pfns < num; |
192 | num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { | |
193 | struct page *page = balloon_page_alloc(); | |
e2250429 | 194 | |
6b35e407 | 195 | if (!page) { |
800ba5ea | 196 | dev_info_ratelimited(&vb->vdev->dev, |
b7dfde95 LT |
197 | "Out of puff! Can't get %u pages\n", |
198 | VIRTIO_BALLOON_PAGES_PER_PAGE); | |
6b35e407 RR |
199 | /* Sleep for at least 1/5 of a second before retry. */ |
200 | msleep(200); | |
201 | break; | |
202 | } | |
c7cdff0e MT |
203 | |
204 | balloon_page_push(&pages, page); | |
205 | } | |
206 | ||
207 | mutex_lock(&vb->balloon_lock); | |
208 | ||
209 | vb->num_pfns = 0; | |
210 | ||
211 | while ((page = balloon_page_pop(&pages))) { | |
212 | balloon_page_enqueue(&vb->vb_dev_info, page); | |
213 | ||
87c9403b | 214 | set_page_pfns(vb, vb->pfns + vb->num_pfns, page); |
3ccc9372 | 215 | vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE; |
997e1208 DL |
216 | if (!virtio_has_feature(vb->vdev, |
217 | VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) | |
218 | adjust_managed_page_count(page, -1); | |
d9e427f6 | 219 | vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE; |
6b35e407 RR |
220 | } |
221 | ||
fad7b7b2 | 222 | num_allocated_pages = vb->num_pfns; |
e2250429 RA |
223 | /* Did we get any? */ |
224 | if (vb->num_pfns != 0) | |
225 | tell_host(vb, vb->inflate_vq); | |
226 | mutex_unlock(&vb->balloon_lock); | |
fad7b7b2 PM |
227 | |
228 | return num_allocated_pages; | |
6b35e407 RR |
229 | } |
230 | ||
195a8c43 LL |
231 | static void release_pages_balloon(struct virtio_balloon *vb, |
232 | struct list_head *pages) | |
6b35e407 | 233 | { |
195a8c43 | 234 | struct page *page, *next; |
6b35e407 | 235 | |
195a8c43 | 236 | list_for_each_entry_safe(page, next, pages, lru) { |
997e1208 DL |
237 | if (!virtio_has_feature(vb->vdev, |
238 | VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) | |
239 | adjust_managed_page_count(page, 1); | |
195a8c43 | 240 | list_del(&page->lru); |
d6d86c0a | 241 | put_page(page); /* balloon reference */ |
6b35e407 RR |
242 | } |
243 | } | |
244 | ||
1fd9c672 | 245 | static unsigned leak_balloon(struct virtio_balloon *vb, size_t num) |
6b35e407 | 246 | { |
1fd9c672 | 247 | unsigned num_freed_pages; |
6b35e407 | 248 | struct page *page; |
9d1ba805 | 249 | struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info; |
195a8c43 | 250 | LIST_HEAD(pages); |
6b35e407 RR |
251 | |
252 | /* We can only do one array worth at a time. */ | |
253 | num = min(num, ARRAY_SIZE(vb->pfns)); | |
254 | ||
e2250429 | 255 | mutex_lock(&vb->balloon_lock); |
37cf99e0 KN |
256 | /* We can't release more pages than taken */ |
257 | num = min(num, (size_t)vb->num_pages); | |
3ccc9372 MT |
258 | for (vb->num_pfns = 0; vb->num_pfns < num; |
259 | vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { | |
e2250429 RA |
260 | page = balloon_page_dequeue(vb_dev_info); |
261 | if (!page) | |
262 | break; | |
87c9403b | 263 | set_page_pfns(vb, vb->pfns + vb->num_pfns, page); |
195a8c43 | 264 | list_add(&page->lru, &pages); |
3ccc9372 | 265 | vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE; |
6b35e407 RR |
266 | } |
267 | ||
1fd9c672 | 268 | num_freed_pages = vb->num_pfns; |
bf50e69f DH |
269 | /* |
270 | * Note that if | |
271 | * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST); | |
272 | * is true, we *have* to do it in this order | |
273 | */ | |
8c6bab4f LC |
274 | if (vb->num_pfns != 0) |
275 | tell_host(vb, vb->deflate_vq); | |
195a8c43 | 276 | release_pages_balloon(vb, &pages); |
f68b992b | 277 | mutex_unlock(&vb->balloon_lock); |
1fd9c672 | 278 | return num_freed_pages; |
6b35e407 RR |
279 | } |
280 | ||
9564e138 AL |
281 | static inline void update_stat(struct virtio_balloon *vb, int idx, |
282 | u16 tag, u64 val) | |
283 | { | |
284 | BUG_ON(idx >= VIRTIO_BALLOON_S_NR); | |
df81b29c MT |
285 | vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag); |
286 | vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val); | |
9564e138 AL |
287 | } |
288 | ||
289 | #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT) | |
290 | ||
9646b26e | 291 | static unsigned int update_balloon_stats(struct virtio_balloon *vb) |
9564e138 AL |
292 | { |
293 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
294 | struct sysinfo i; | |
9646b26e | 295 | unsigned int idx = 0; |
5057dcd0 | 296 | long available; |
4d32029b | 297 | unsigned long caches; |
9564e138 AL |
298 | |
299 | all_vm_events(events); | |
300 | si_meminfo(&i); | |
301 | ||
5057dcd0 | 302 | available = si_mem_available(); |
4d32029b | 303 | caches = global_node_page_state(NR_FILE_PAGES); |
5057dcd0 | 304 | |
f0bb2d50 | 305 | #ifdef CONFIG_VM_EVENT_COUNTERS |
9564e138 AL |
306 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN, |
307 | pages_to_bytes(events[PSWPIN])); | |
308 | update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT, | |
309 | pages_to_bytes(events[PSWPOUT])); | |
310 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]); | |
311 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]); | |
6c64fe7f JH |
312 | #ifdef CONFIG_HUGETLB_PAGE |
313 | update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC, | |
314 | events[HTLB_BUDDY_PGALLOC]); | |
315 | update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL, | |
316 | events[HTLB_BUDDY_PGALLOC_FAIL]); | |
317 | #endif | |
f0bb2d50 | 318 | #endif |
9564e138 AL |
319 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE, |
320 | pages_to_bytes(i.freeram)); | |
321 | update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT, | |
322 | pages_to_bytes(i.totalram)); | |
5057dcd0 IR |
323 | update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL, |
324 | pages_to_bytes(available)); | |
4d32029b TG |
325 | update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES, |
326 | pages_to_bytes(caches)); | |
9646b26e LP |
327 | |
328 | return idx; | |
9564e138 AL |
329 | } |
330 | ||
331 | /* | |
332 | * While most virtqueues communicate guest-initiated requests to the hypervisor, | |
333 | * the stats queue operates in reverse. The driver initializes the virtqueue | |
334 | * with a single buffer. From that point forward, all conversations consist of | |
335 | * a hypervisor request (a call to this function) which directs us to refill | |
1f34c71a | 336 | * the virtqueue with a fresh stats buffer. Since stats collection can sleep, |
fad7b7b2 PM |
337 | * we delegate the job to a freezable workqueue that will do the actual work via |
338 | * stats_handle_request(). | |
9564e138 | 339 | */ |
1f34c71a | 340 | static void stats_request(struct virtqueue *vq) |
9564e138 | 341 | { |
9c378abc | 342 | struct virtio_balloon *vb = vq->vdev->priv; |
9564e138 | 343 | |
fad7b7b2 PM |
344 | spin_lock(&vb->stop_update_lock); |
345 | if (!vb->stop_update) | |
fd0e21c3 | 346 | queue_work(system_freezable_wq, &vb->update_balloon_stats_work); |
fad7b7b2 | 347 | spin_unlock(&vb->stop_update_lock); |
1f34c71a AL |
348 | } |
349 | ||
350 | static void stats_handle_request(struct virtio_balloon *vb) | |
351 | { | |
352 | struct virtqueue *vq; | |
353 | struct scatterlist sg; | |
9646b26e | 354 | unsigned int len, num_stats; |
9564e138 | 355 | |
9646b26e | 356 | num_stats = update_balloon_stats(vb); |
9564e138 | 357 | |
1f34c71a | 358 | vq = vb->stats_vq; |
9c378abc MT |
359 | if (!virtqueue_get_buf(vq, &len)) |
360 | return; | |
9646b26e | 361 | sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats); |
4951cc90 | 362 | virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); |
946cfe0e | 363 | virtqueue_kick(vq); |
9564e138 AL |
364 | } |
365 | ||
bdc1681c | 366 | static inline s64 towards_target(struct virtio_balloon *vb) |
6b35e407 | 367 | { |
1a87228f | 368 | s64 target; |
df81b29c | 369 | u32 num_pages; |
1a87228f | 370 | |
df81b29c MT |
371 | virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages, |
372 | &num_pages); | |
855e0c52 | 373 | |
df81b29c MT |
374 | /* Legacy balloon config space is LE, unlike all other devices. */ |
375 | if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1)) | |
376 | num_pages = le32_to_cpu((__force __le32)num_pages); | |
377 | ||
378 | target = num_pages; | |
1a87228f | 379 | return target - vb->num_pages; |
6b35e407 RR |
380 | } |
381 | ||
86a55978 WW |
382 | /* Gives back @num_to_return blocks of free pages to mm. */ |
383 | static unsigned long return_free_pages_to_mm(struct virtio_balloon *vb, | |
384 | unsigned long num_to_return) | |
385 | { | |
386 | struct page *page; | |
387 | unsigned long num_returned; | |
388 | ||
389 | spin_lock_irq(&vb->free_page_list_lock); | |
390 | for (num_returned = 0; num_returned < num_to_return; num_returned++) { | |
391 | page = balloon_page_pop(&vb->free_page_list); | |
392 | if (!page) | |
393 | break; | |
394 | free_pages((unsigned long)page_address(page), | |
395 | VIRTIO_BALLOON_FREE_PAGE_ORDER); | |
396 | } | |
397 | vb->num_free_page_blocks -= num_returned; | |
398 | spin_unlock_irq(&vb->free_page_list_lock); | |
399 | ||
400 | return num_returned; | |
401 | } | |
402 | ||
bf4dc0b2 WW |
403 | static void virtio_balloon_queue_free_page_work(struct virtio_balloon *vb) |
404 | { | |
405 | if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) | |
406 | return; | |
407 | ||
408 | /* No need to queue the work if the bit was already set. */ | |
409 | if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID, | |
410 | &vb->config_read_bitmap)) | |
411 | return; | |
412 | ||
413 | queue_work(vb->balloon_wq, &vb->report_free_page_work); | |
414 | } | |
415 | ||
86a55978 WW |
416 | static void virtballoon_changed(struct virtio_device *vdev) |
417 | { | |
418 | struct virtio_balloon *vb = vdev->priv; | |
419 | unsigned long flags; | |
86a55978 | 420 | |
bf4dc0b2 WW |
421 | spin_lock_irqsave(&vb->stop_update_lock, flags); |
422 | if (!vb->stop_update) { | |
423 | queue_work(system_freezable_wq, | |
424 | &vb->update_balloon_size_work); | |
425 | virtio_balloon_queue_free_page_work(vb); | |
86a55978 | 426 | } |
bf4dc0b2 | 427 | spin_unlock_irqrestore(&vb->stop_update_lock, flags); |
86a55978 WW |
428 | } |
429 | ||
6b35e407 RR |
430 | static void update_balloon_size(struct virtio_balloon *vb) |
431 | { | |
df81b29c MT |
432 | u32 actual = vb->num_pages; |
433 | ||
434 | /* Legacy balloon config space is LE, unlike all other devices. */ | |
435 | if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1)) | |
436 | actual = (__force u32)cpu_to_le32(actual); | |
6b35e407 | 437 | |
3459f11a | 438 | virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual, |
855e0c52 | 439 | &actual); |
6b35e407 RR |
440 | } |
441 | ||
fd0e21c3 | 442 | static void update_balloon_stats_func(struct work_struct *work) |
6b35e407 | 443 | { |
fd0e21c3 | 444 | struct virtio_balloon *vb; |
3d2a3774 | 445 | |
fd0e21c3 PM |
446 | vb = container_of(work, struct virtio_balloon, |
447 | update_balloon_stats_work); | |
448 | stats_handle_request(vb); | |
449 | } | |
1f74ef0f | 450 | |
fd0e21c3 | 451 | static void update_balloon_size_func(struct work_struct *work) |
6b35e407 | 452 | { |
fad7b7b2 PM |
453 | struct virtio_balloon *vb; |
454 | s64 diff; | |
3d2a3774 | 455 | |
fd0e21c3 PM |
456 | vb = container_of(work, struct virtio_balloon, |
457 | update_balloon_size_work); | |
fad7b7b2 | 458 | diff = towards_target(vb); |
1f74ef0f | 459 | |
53e946cb WW |
460 | if (!diff) |
461 | return; | |
462 | ||
fad7b7b2 PM |
463 | if (diff > 0) |
464 | diff -= fill_balloon(vb, diff); | |
53e946cb | 465 | else |
fad7b7b2 PM |
466 | diff += leak_balloon(vb, -diff); |
467 | update_balloon_size(vb); | |
468 | ||
469 | if (diff) | |
470 | queue_work(system_freezable_wq, work); | |
6b35e407 RR |
471 | } |
472 | ||
be91c33d | 473 | static int init_vqs(struct virtio_balloon *vb) |
6b35e407 | 474 | { |
86a55978 WW |
475 | struct virtqueue *vqs[VIRTIO_BALLOON_VQ_MAX]; |
476 | vq_callback_t *callbacks[VIRTIO_BALLOON_VQ_MAX]; | |
477 | const char *names[VIRTIO_BALLOON_VQ_MAX]; | |
478 | int err; | |
6b35e407 | 479 | |
be91c33d | 480 | /* |
86a55978 WW |
481 | * Inflateq and deflateq are used unconditionally. The names[] |
482 | * will be NULL if the related feature is not enabled, which will | |
483 | * cause no allocation for the corresponding virtqueue in find_vqs. | |
be91c33d | 484 | */ |
86a55978 WW |
485 | callbacks[VIRTIO_BALLOON_VQ_INFLATE] = balloon_ack; |
486 | names[VIRTIO_BALLOON_VQ_INFLATE] = "inflate"; | |
487 | callbacks[VIRTIO_BALLOON_VQ_DEFLATE] = balloon_ack; | |
488 | names[VIRTIO_BALLOON_VQ_DEFLATE] = "deflate"; | |
489 | names[VIRTIO_BALLOON_VQ_STATS] = NULL; | |
490 | names[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL; | |
491 | ||
492 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { | |
493 | names[VIRTIO_BALLOON_VQ_STATS] = "stats"; | |
494 | callbacks[VIRTIO_BALLOON_VQ_STATS] = stats_request; | |
495 | } | |
496 | ||
497 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { | |
498 | names[VIRTIO_BALLOON_VQ_FREE_PAGE] = "free_page_vq"; | |
499 | callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL; | |
500 | } | |
501 | ||
502 | err = vb->vdev->config->find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX, | |
503 | vqs, callbacks, names, NULL, NULL); | |
d2a7ddda | 504 | if (err) |
be91c33d | 505 | return err; |
6b35e407 | 506 | |
86a55978 WW |
507 | vb->inflate_vq = vqs[VIRTIO_BALLOON_VQ_INFLATE]; |
508 | vb->deflate_vq = vqs[VIRTIO_BALLOON_VQ_DEFLATE]; | |
9564e138 AL |
509 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { |
510 | struct scatterlist sg; | |
9646b26e | 511 | unsigned int num_stats; |
86a55978 | 512 | vb->stats_vq = vqs[VIRTIO_BALLOON_VQ_STATS]; |
9564e138 AL |
513 | |
514 | /* | |
515 | * Prime this virtqueue with one buffer so the hypervisor can | |
4951cc90 | 516 | * use it to signal us later (it can't be broken yet!). |
9564e138 | 517 | */ |
9646b26e | 518 | num_stats = update_balloon_stats(vb); |
fc865322 | 519 | |
9646b26e | 520 | sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats); |
74cf5b16 WW |
521 | err = virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, |
522 | GFP_KERNEL); | |
523 | if (err) { | |
524 | dev_warn(&vb->vdev->dev, "%s: add stat_vq failed\n", | |
525 | __func__); | |
526 | return err; | |
527 | } | |
946cfe0e | 528 | virtqueue_kick(vb->stats_vq); |
9564e138 | 529 | } |
86a55978 WW |
530 | |
531 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) | |
532 | vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE]; | |
533 | ||
534 | return 0; | |
535 | } | |
536 | ||
bf4dc0b2 WW |
537 | static u32 virtio_balloon_cmd_id_received(struct virtio_balloon *vb) |
538 | { | |
539 | if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID, | |
540 | &vb->config_read_bitmap)) | |
541 | virtio_cread(vb->vdev, struct virtio_balloon_config, | |
542 | free_page_report_cmd_id, | |
543 | &vb->cmd_id_received_cache); | |
544 | ||
545 | return vb->cmd_id_received_cache; | |
546 | } | |
547 | ||
86a55978 WW |
548 | static int send_cmd_id_start(struct virtio_balloon *vb) |
549 | { | |
550 | struct scatterlist sg; | |
551 | struct virtqueue *vq = vb->free_page_vq; | |
552 | int err, unused; | |
553 | ||
554 | /* Detach all the used buffers from the vq */ | |
555 | while (virtqueue_get_buf(vq, &unused)) | |
556 | ; | |
557 | ||
bf4dc0b2 WW |
558 | vb->cmd_id_active = virtio32_to_cpu(vb->vdev, |
559 | virtio_balloon_cmd_id_received(vb)); | |
86a55978 WW |
560 | sg_init_one(&sg, &vb->cmd_id_active, sizeof(vb->cmd_id_active)); |
561 | err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_active, GFP_KERNEL); | |
562 | if (!err) | |
563 | virtqueue_kick(vq); | |
564 | return err; | |
565 | } | |
566 | ||
567 | static int send_cmd_id_stop(struct virtio_balloon *vb) | |
568 | { | |
569 | struct scatterlist sg; | |
570 | struct virtqueue *vq = vb->free_page_vq; | |
571 | int err, unused; | |
572 | ||
573 | /* Detach all the used buffers from the vq */ | |
574 | while (virtqueue_get_buf(vq, &unused)) | |
575 | ; | |
576 | ||
577 | sg_init_one(&sg, &vb->cmd_id_stop, sizeof(vb->cmd_id_stop)); | |
578 | err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_stop, GFP_KERNEL); | |
579 | if (!err) | |
580 | virtqueue_kick(vq); | |
581 | return err; | |
582 | } | |
583 | ||
584 | static int get_free_page_and_send(struct virtio_balloon *vb) | |
585 | { | |
586 | struct virtqueue *vq = vb->free_page_vq; | |
587 | struct page *page; | |
588 | struct scatterlist sg; | |
589 | int err, unused; | |
590 | void *p; | |
591 | ||
592 | /* Detach all the used buffers from the vq */ | |
593 | while (virtqueue_get_buf(vq, &unused)) | |
594 | ; | |
595 | ||
596 | page = alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG, | |
597 | VIRTIO_BALLOON_FREE_PAGE_ORDER); | |
598 | /* | |
599 | * When the allocation returns NULL, it indicates that we have got all | |
600 | * the possible free pages, so return -EINTR to stop. | |
601 | */ | |
602 | if (!page) | |
603 | return -EINTR; | |
604 | ||
605 | p = page_address(page); | |
606 | sg_init_one(&sg, p, VIRTIO_BALLOON_FREE_PAGE_SIZE); | |
607 | /* There is always 1 entry reserved for the cmd id to use. */ | |
608 | if (vq->num_free > 1) { | |
609 | err = virtqueue_add_inbuf(vq, &sg, 1, p, GFP_KERNEL); | |
610 | if (unlikely(err)) { | |
611 | free_pages((unsigned long)p, | |
612 | VIRTIO_BALLOON_FREE_PAGE_ORDER); | |
613 | return err; | |
614 | } | |
615 | virtqueue_kick(vq); | |
616 | spin_lock_irq(&vb->free_page_list_lock); | |
617 | balloon_page_push(&vb->free_page_list, page); | |
618 | vb->num_free_page_blocks++; | |
619 | spin_unlock_irq(&vb->free_page_list_lock); | |
620 | } else { | |
621 | /* | |
622 | * The vq has no available entry to add this page block, so | |
623 | * just free it. | |
624 | */ | |
625 | free_pages((unsigned long)p, VIRTIO_BALLOON_FREE_PAGE_ORDER); | |
626 | } | |
627 | ||
be91c33d AS |
628 | return 0; |
629 | } | |
630 | ||
86a55978 WW |
631 | static int send_free_pages(struct virtio_balloon *vb) |
632 | { | |
633 | int err; | |
634 | u32 cmd_id_active; | |
635 | ||
636 | while (1) { | |
637 | /* | |
638 | * If a stop id or a new cmd id was just received from host, | |
639 | * stop the reporting. | |
640 | */ | |
641 | cmd_id_active = virtio32_to_cpu(vb->vdev, vb->cmd_id_active); | |
bf4dc0b2 WW |
642 | if (unlikely(cmd_id_active != |
643 | virtio_balloon_cmd_id_received(vb))) | |
86a55978 WW |
644 | break; |
645 | ||
646 | /* | |
647 | * The free page blocks are allocated and sent to host one by | |
648 | * one. | |
649 | */ | |
650 | err = get_free_page_and_send(vb); | |
651 | if (err == -EINTR) | |
652 | break; | |
653 | else if (unlikely(err)) | |
654 | return err; | |
655 | } | |
656 | ||
657 | return 0; | |
658 | } | |
659 | ||
bf4dc0b2 | 660 | static void virtio_balloon_report_free_page(struct virtio_balloon *vb) |
86a55978 WW |
661 | { |
662 | int err; | |
86a55978 WW |
663 | struct device *dev = &vb->vdev->dev; |
664 | ||
665 | /* Start by sending the received cmd id to host with an outbuf. */ | |
666 | err = send_cmd_id_start(vb); | |
667 | if (unlikely(err)) | |
668 | dev_err(dev, "Failed to send a start id, err = %d\n", err); | |
669 | ||
670 | err = send_free_pages(vb); | |
671 | if (unlikely(err)) | |
672 | dev_err(dev, "Failed to send a free page, err = %d\n", err); | |
673 | ||
674 | /* End by sending a stop id to host with an outbuf. */ | |
675 | err = send_cmd_id_stop(vb); | |
676 | if (unlikely(err)) | |
677 | dev_err(dev, "Failed to send a stop id, err = %d\n", err); | |
678 | } | |
679 | ||
bf4dc0b2 WW |
680 | static void report_free_page_func(struct work_struct *work) |
681 | { | |
682 | struct virtio_balloon *vb = container_of(work, struct virtio_balloon, | |
683 | report_free_page_work); | |
684 | u32 cmd_id_received; | |
685 | ||
686 | cmd_id_received = virtio_balloon_cmd_id_received(vb); | |
687 | if (cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) { | |
688 | /* Pass ULONG_MAX to give back all the free pages */ | |
689 | return_free_pages_to_mm(vb, ULONG_MAX); | |
690 | } else if (cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP && | |
691 | cmd_id_received != | |
692 | virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) { | |
693 | virtio_balloon_report_free_page(vb); | |
694 | } | |
695 | } | |
696 | ||
e2250429 RA |
697 | #ifdef CONFIG_BALLOON_COMPACTION |
698 | /* | |
699 | * virtballoon_migratepage - perform the balloon page migration on behalf of | |
700 | * a compation thread. (called under page lock) | |
9d1ba805 | 701 | * @vb_dev_info: the balloon device |
e2250429 RA |
702 | * @newpage: page that will replace the isolated page after migration finishes. |
703 | * @page : the isolated (old) page that is about to be migrated to newpage. | |
704 | * @mode : compaction mode -- not used for balloon page migration. | |
705 | * | |
706 | * After a ballooned page gets isolated by compaction procedures, this is the | |
707 | * function that performs the page migration on behalf of a compaction thread | |
708 | * The page migration for virtio balloon is done in a simple swap fashion which | |
709 | * follows these two macro steps: | |
710 | * 1) insert newpage into vb->pages list and update the host about it; | |
711 | * 2) update the host about the old page removed from vb->pages list; | |
712 | * | |
713 | * This function preforms the balloon page migration task. | |
714 | * Called through balloon_mapping->a_ops->migratepage | |
715 | */ | |
9d1ba805 | 716 | static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info, |
e2250429 RA |
717 | struct page *newpage, struct page *page, enum migrate_mode mode) |
718 | { | |
9d1ba805 KK |
719 | struct virtio_balloon *vb = container_of(vb_dev_info, |
720 | struct virtio_balloon, vb_dev_info); | |
e2250429 RA |
721 | unsigned long flags; |
722 | ||
e2250429 RA |
723 | /* |
724 | * In order to avoid lock contention while migrating pages concurrently | |
725 | * to leak_balloon() or fill_balloon() we just give up the balloon_lock | |
726 | * this turn, as it is easier to retry the page migration later. | |
727 | * This also prevents fill_balloon() getting stuck into a mutex | |
728 | * recursion in the case it ends up triggering memory compaction | |
729 | * while it is attempting to inflate the ballon. | |
730 | */ | |
731 | if (!mutex_trylock(&vb->balloon_lock)) | |
732 | return -EAGAIN; | |
733 | ||
d6d86c0a KK |
734 | get_page(newpage); /* balloon reference */ |
735 | ||
e2250429 RA |
736 | /* balloon's page migration 1st step -- inflate "newpage" */ |
737 | spin_lock_irqsave(&vb_dev_info->pages_lock, flags); | |
9d1ba805 | 738 | balloon_page_insert(vb_dev_info, newpage); |
e2250429 | 739 | vb_dev_info->isolated_pages--; |
09316c09 | 740 | __count_vm_event(BALLOON_MIGRATE); |
e2250429 RA |
741 | spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags); |
742 | vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; | |
87c9403b | 743 | set_page_pfns(vb, vb->pfns, newpage); |
e2250429 RA |
744 | tell_host(vb, vb->inflate_vq); |
745 | ||
d6d86c0a | 746 | /* balloon's page migration 2nd step -- deflate "page" */ |
89da619b | 747 | spin_lock_irqsave(&vb_dev_info->pages_lock, flags); |
e2250429 | 748 | balloon_page_delete(page); |
89da619b | 749 | spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags); |
e2250429 | 750 | vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; |
87c9403b | 751 | set_page_pfns(vb, vb->pfns, page); |
e2250429 RA |
752 | tell_host(vb, vb->deflate_vq); |
753 | ||
754 | mutex_unlock(&vb->balloon_lock); | |
755 | ||
d6d86c0a KK |
756 | put_page(page); /* balloon reference */ |
757 | ||
dd4123f3 | 758 | return MIGRATEPAGE_SUCCESS; |
e2250429 | 759 | } |
b1123ea6 MK |
760 | |
761 | static struct dentry *balloon_mount(struct file_system_type *fs_type, | |
762 | int flags, const char *dev_name, void *data) | |
763 | { | |
1f58bb18 | 764 | return mount_pseudo(fs_type, NULL, NULL, BALLOON_KVM_MAGIC); |
b1123ea6 MK |
765 | } |
766 | ||
767 | static struct file_system_type balloon_fs = { | |
768 | .name = "balloon-kvm", | |
769 | .mount = balloon_mount, | |
770 | .kill_sb = kill_anon_super, | |
771 | }; | |
772 | ||
e2250429 RA |
773 | #endif /* CONFIG_BALLOON_COMPACTION */ |
774 | ||
86a55978 WW |
775 | static unsigned long shrink_free_pages(struct virtio_balloon *vb, |
776 | unsigned long pages_to_free) | |
71994620 | 777 | { |
86a55978 | 778 | unsigned long blocks_to_free, blocks_freed; |
71994620 | 779 | |
86a55978 WW |
780 | pages_to_free = round_up(pages_to_free, |
781 | 1 << VIRTIO_BALLOON_FREE_PAGE_ORDER); | |
782 | blocks_to_free = pages_to_free >> VIRTIO_BALLOON_FREE_PAGE_ORDER; | |
783 | blocks_freed = return_free_pages_to_mm(vb, blocks_to_free); | |
784 | ||
785 | return blocks_freed << VIRTIO_BALLOON_FREE_PAGE_ORDER; | |
786 | } | |
787 | ||
788 | static unsigned long shrink_balloon_pages(struct virtio_balloon *vb, | |
789 | unsigned long pages_to_free) | |
790 | { | |
791 | unsigned long pages_freed = 0; | |
71994620 WW |
792 | |
793 | /* | |
794 | * One invocation of leak_balloon can deflate at most | |
795 | * VIRTIO_BALLOON_ARRAY_PFNS_MAX balloon pages, so we call it | |
796 | * multiple times to deflate pages till reaching pages_to_free. | |
797 | */ | |
798 | while (vb->num_pages && pages_to_free) { | |
86a55978 WW |
799 | pages_freed += leak_balloon(vb, pages_to_free) / |
800 | VIRTIO_BALLOON_PAGES_PER_PAGE; | |
71994620 | 801 | pages_to_free -= pages_freed; |
71994620 WW |
802 | } |
803 | update_balloon_size(vb); | |
804 | ||
86a55978 WW |
805 | return pages_freed; |
806 | } | |
807 | ||
808 | static unsigned long virtio_balloon_shrinker_scan(struct shrinker *shrinker, | |
809 | struct shrink_control *sc) | |
810 | { | |
811 | unsigned long pages_to_free, pages_freed = 0; | |
812 | struct virtio_balloon *vb = container_of(shrinker, | |
813 | struct virtio_balloon, shrinker); | |
814 | ||
815 | pages_to_free = sc->nr_to_scan * VIRTIO_BALLOON_PAGES_PER_PAGE; | |
816 | ||
817 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) | |
818 | pages_freed = shrink_free_pages(vb, pages_to_free); | |
819 | ||
820 | if (pages_freed >= pages_to_free) | |
821 | return pages_freed; | |
822 | ||
823 | pages_freed += shrink_balloon_pages(vb, pages_to_free - pages_freed); | |
824 | ||
825 | return pages_freed; | |
71994620 WW |
826 | } |
827 | ||
828 | static unsigned long virtio_balloon_shrinker_count(struct shrinker *shrinker, | |
829 | struct shrink_control *sc) | |
830 | { | |
831 | struct virtio_balloon *vb = container_of(shrinker, | |
832 | struct virtio_balloon, shrinker); | |
86a55978 | 833 | unsigned long count; |
71994620 | 834 | |
86a55978 WW |
835 | count = vb->num_pages / VIRTIO_BALLOON_PAGES_PER_PAGE; |
836 | count += vb->num_free_page_blocks >> VIRTIO_BALLOON_FREE_PAGE_ORDER; | |
837 | ||
838 | return count; | |
71994620 WW |
839 | } |
840 | ||
841 | static void virtio_balloon_unregister_shrinker(struct virtio_balloon *vb) | |
842 | { | |
843 | unregister_shrinker(&vb->shrinker); | |
844 | } | |
845 | ||
846 | static int virtio_balloon_register_shrinker(struct virtio_balloon *vb) | |
847 | { | |
848 | vb->shrinker.scan_objects = virtio_balloon_shrinker_scan; | |
849 | vb->shrinker.count_objects = virtio_balloon_shrinker_count; | |
850 | vb->shrinker.seeks = DEFAULT_SEEKS; | |
851 | ||
852 | return register_shrinker(&vb->shrinker); | |
853 | } | |
854 | ||
be91c33d AS |
855 | static int virtballoon_probe(struct virtio_device *vdev) |
856 | { | |
857 | struct virtio_balloon *vb; | |
2e991629 | 858 | __u32 poison_val; |
be91c33d AS |
859 | int err; |
860 | ||
2d9becc1 MT |
861 | if (!vdev->config->get) { |
862 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
863 | __func__); | |
864 | return -EINVAL; | |
865 | } | |
866 | ||
c51d8fca | 867 | vdev->priv = vb = kzalloc(sizeof(*vb), GFP_KERNEL); |
be91c33d AS |
868 | if (!vb) { |
869 | err = -ENOMEM; | |
870 | goto out; | |
871 | } | |
872 | ||
fd0e21c3 PM |
873 | INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func); |
874 | INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func); | |
fad7b7b2 | 875 | spin_lock_init(&vb->stop_update_lock); |
e2250429 | 876 | mutex_init(&vb->balloon_lock); |
9c378abc | 877 | init_waitqueue_head(&vb->acked); |
be91c33d | 878 | vb->vdev = vdev; |
be91c33d | 879 | |
9d1ba805 | 880 | balloon_devinfo_init(&vb->vb_dev_info); |
e2250429 | 881 | |
be91c33d AS |
882 | err = init_vqs(vb); |
883 | if (err) | |
9d1ba805 | 884 | goto out_free_vb; |
6b35e407 | 885 | |
b1123ea6 MK |
886 | #ifdef CONFIG_BALLOON_COMPACTION |
887 | balloon_mnt = kern_mount(&balloon_fs); | |
888 | if (IS_ERR(balloon_mnt)) { | |
889 | err = PTR_ERR(balloon_mnt); | |
b1123ea6 MK |
890 | goto out_del_vqs; |
891 | } | |
892 | ||
893 | vb->vb_dev_info.migratepage = virtballoon_migratepage; | |
894 | vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb); | |
895 | if (IS_ERR(vb->vb_dev_info.inode)) { | |
896 | err = PTR_ERR(vb->vb_dev_info.inode); | |
897 | kern_unmount(balloon_mnt); | |
b1123ea6 MK |
898 | goto out_del_vqs; |
899 | } | |
900 | vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops; | |
901 | #endif | |
86a55978 WW |
902 | if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { |
903 | /* | |
904 | * There is always one entry reserved for cmd id, so the ring | |
905 | * size needs to be at least two to report free page hints. | |
906 | */ | |
907 | if (virtqueue_get_vring_size(vb->free_page_vq) < 2) { | |
908 | err = -ENOSPC; | |
909 | goto out_del_vqs; | |
910 | } | |
911 | vb->balloon_wq = alloc_workqueue("balloon-wq", | |
912 | WQ_FREEZABLE | WQ_CPU_INTENSIVE, 0); | |
913 | if (!vb->balloon_wq) { | |
914 | err = -ENOMEM; | |
915 | goto out_del_vqs; | |
916 | } | |
917 | INIT_WORK(&vb->report_free_page_work, report_free_page_func); | |
bf4dc0b2 | 918 | vb->cmd_id_received_cache = VIRTIO_BALLOON_CMD_ID_STOP; |
86a55978 WW |
919 | vb->cmd_id_active = cpu_to_virtio32(vb->vdev, |
920 | VIRTIO_BALLOON_CMD_ID_STOP); | |
921 | vb->cmd_id_stop = cpu_to_virtio32(vb->vdev, | |
922 | VIRTIO_BALLOON_CMD_ID_STOP); | |
86a55978 WW |
923 | spin_lock_init(&vb->free_page_list_lock); |
924 | INIT_LIST_HEAD(&vb->free_page_list); | |
2e991629 WW |
925 | if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) { |
926 | memset(&poison_val, PAGE_POISON, sizeof(poison_val)); | |
927 | virtio_cwrite(vb->vdev, struct virtio_balloon_config, | |
928 | poison_val, &poison_val); | |
929 | } | |
86a55978 | 930 | } |
71994620 WW |
931 | /* |
932 | * We continue to use VIRTIO_BALLOON_F_DEFLATE_ON_OOM to decide if a | |
933 | * shrinker needs to be registered to relieve memory pressure. | |
934 | */ | |
935 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) { | |
936 | err = virtio_balloon_register_shrinker(vb); | |
937 | if (err) | |
86a55978 | 938 | goto out_del_balloon_wq; |
71994620 | 939 | } |
88660f7f MT |
940 | virtio_device_ready(vdev); |
941 | ||
8424af53 KN |
942 | if (towards_target(vb)) |
943 | virtballoon_changed(vdev); | |
6b35e407 RR |
944 | return 0; |
945 | ||
86a55978 WW |
946 | out_del_balloon_wq: |
947 | if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) | |
948 | destroy_workqueue(vb->balloon_wq); | |
b1123ea6 | 949 | out_del_vqs: |
d2a7ddda | 950 | vdev->config->del_vqs(vdev); |
6b35e407 RR |
951 | out_free_vb: |
952 | kfree(vb); | |
953 | out: | |
954 | return err; | |
955 | } | |
956 | ||
c877bab5 | 957 | static void remove_common(struct virtio_balloon *vb) |
6b35e407 | 958 | { |
6b35e407 RR |
959 | /* There might be pages left in the balloon: free them. */ |
960 | while (vb->num_pages) | |
961 | leak_balloon(vb, vb->num_pages); | |
b8ae0eb3 | 962 | update_balloon_size(vb); |
6b35e407 RR |
963 | |
964 | /* Now we reset the device so we can clean up the queues. */ | |
c877bab5 | 965 | vb->vdev->config->reset(vb->vdev); |
6b35e407 | 966 | |
c877bab5 AS |
967 | vb->vdev->config->del_vqs(vb->vdev); |
968 | } | |
969 | ||
8590dbc7 | 970 | static void virtballoon_remove(struct virtio_device *vdev) |
c877bab5 AS |
971 | { |
972 | struct virtio_balloon *vb = vdev->priv; | |
973 | ||
71994620 WW |
974 | if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) |
975 | virtio_balloon_unregister_shrinker(vb); | |
fad7b7b2 PM |
976 | spin_lock_irq(&vb->stop_update_lock); |
977 | vb->stop_update = true; | |
978 | spin_unlock_irq(&vb->stop_update_lock); | |
fd0e21c3 PM |
979 | cancel_work_sync(&vb->update_balloon_size_work); |
980 | cancel_work_sync(&vb->update_balloon_stats_work); | |
fad7b7b2 | 981 | |
86a55978 WW |
982 | if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { |
983 | cancel_work_sync(&vb->report_free_page_work); | |
984 | destroy_workqueue(vb->balloon_wq); | |
985 | } | |
986 | ||
c877bab5 | 987 | remove_common(vb); |
9c57b580 | 988 | #ifdef CONFIG_BALLOON_COMPACTION |
b1123ea6 MK |
989 | if (vb->vb_dev_info.inode) |
990 | iput(vb->vb_dev_info.inode); | |
9c57b580 YX |
991 | |
992 | kern_unmount(balloon_mnt); | |
993 | #endif | |
6b35e407 RR |
994 | kfree(vb); |
995 | } | |
996 | ||
89107000 | 997 | #ifdef CONFIG_PM_SLEEP |
e562966d AS |
998 | static int virtballoon_freeze(struct virtio_device *vdev) |
999 | { | |
4eb05d56 AS |
1000 | struct virtio_balloon *vb = vdev->priv; |
1001 | ||
e562966d | 1002 | /* |
fad7b7b2 | 1003 | * The workqueue is already frozen by the PM core before this |
e562966d AS |
1004 | * function is called. |
1005 | */ | |
c877bab5 | 1006 | remove_common(vb); |
e562966d AS |
1007 | return 0; |
1008 | } | |
1009 | ||
c45b4166 | 1010 | static int virtballoon_restore(struct virtio_device *vdev) |
4eb05d56 AS |
1011 | { |
1012 | struct virtio_balloon *vb = vdev->priv; | |
1013 | int ret; | |
1014 | ||
1015 | ret = init_vqs(vdev->priv); | |
1016 | if (ret) | |
1017 | return ret; | |
1018 | ||
486d2e63 MT |
1019 | virtio_device_ready(vdev); |
1020 | ||
fad7b7b2 PM |
1021 | if (towards_target(vb)) |
1022 | virtballoon_changed(vdev); | |
4eb05d56 AS |
1023 | update_balloon_size(vb); |
1024 | return 0; | |
1025 | } | |
e562966d AS |
1026 | #endif |
1027 | ||
e41b1355 MT |
1028 | static int virtballoon_validate(struct virtio_device *vdev) |
1029 | { | |
2e991629 WW |
1030 | if (!page_poisoning_enabled()) |
1031 | __virtio_clear_bit(vdev, VIRTIO_BALLOON_F_PAGE_POISON); | |
1032 | ||
e41b1355 MT |
1033 | __virtio_clear_bit(vdev, VIRTIO_F_IOMMU_PLATFORM); |
1034 | return 0; | |
1035 | } | |
1036 | ||
9564e138 AL |
1037 | static unsigned int features[] = { |
1038 | VIRTIO_BALLOON_F_MUST_TELL_HOST, | |
1039 | VIRTIO_BALLOON_F_STATS_VQ, | |
5a10b7db | 1040 | VIRTIO_BALLOON_F_DEFLATE_ON_OOM, |
86a55978 | 1041 | VIRTIO_BALLOON_F_FREE_PAGE_HINT, |
2e991629 | 1042 | VIRTIO_BALLOON_F_PAGE_POISON, |
9564e138 | 1043 | }; |
c45a6816 | 1044 | |
d817cd52 | 1045 | static struct virtio_driver virtio_balloon_driver = { |
c45a6816 RR |
1046 | .feature_table = features, |
1047 | .feature_table_size = ARRAY_SIZE(features), | |
6b35e407 RR |
1048 | .driver.name = KBUILD_MODNAME, |
1049 | .driver.owner = THIS_MODULE, | |
1050 | .id_table = id_table, | |
e41b1355 | 1051 | .validate = virtballoon_validate, |
6b35e407 | 1052 | .probe = virtballoon_probe, |
8590dbc7 | 1053 | .remove = virtballoon_remove, |
6b35e407 | 1054 | .config_changed = virtballoon_changed, |
89107000 | 1055 | #ifdef CONFIG_PM_SLEEP |
e562966d AS |
1056 | .freeze = virtballoon_freeze, |
1057 | .restore = virtballoon_restore, | |
e562966d | 1058 | #endif |
6b35e407 RR |
1059 | }; |
1060 | ||
b2a17029 | 1061 | module_virtio_driver(virtio_balloon_driver); |
6b35e407 RR |
1062 | MODULE_DEVICE_TABLE(virtio, id_table); |
1063 | MODULE_DESCRIPTION("Virtio balloon driver"); | |
1064 | MODULE_LICENSE("GPL"); |