Loading...
1/* Virtio balloon implementation, inspired by Dor Loar and Marcelo
2 * Tosatti's implementations.
3 *
4 * Copyright 2008 Rusty Russell IBM Corporation
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20//#define DEBUG
21#include <linux/virtio.h>
22#include <linux/virtio_balloon.h>
23#include <linux/swap.h>
24#include <linux/kthread.h>
25#include <linux/freezer.h>
26#include <linux/delay.h>
27#include <linux/slab.h>
28
29struct virtio_balloon
30{
31 struct virtio_device *vdev;
32 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
33
34 /* Where the ballooning thread waits for config to change. */
35 wait_queue_head_t config_change;
36
37 /* The thread servicing the balloon. */
38 struct task_struct *thread;
39
40 /* Waiting for host to ack the pages we released. */
41 struct completion acked;
42
43 /* The pages we've told the Host we're not using. */
44 unsigned int num_pages;
45 struct list_head pages;
46
47 /* The array of pfns we tell the Host about. */
48 unsigned int num_pfns;
49 u32 pfns[256];
50
51 /* Memory statistics */
52 int need_stats_update;
53 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
54};
55
56static struct virtio_device_id id_table[] = {
57 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
58 { 0 },
59};
60
61static u32 page_to_balloon_pfn(struct page *page)
62{
63 unsigned long pfn = page_to_pfn(page);
64
65 BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
66 /* Convert pfn from Linux page size to balloon page size. */
67 return pfn >> (PAGE_SHIFT - VIRTIO_BALLOON_PFN_SHIFT);
68}
69
70static void balloon_ack(struct virtqueue *vq)
71{
72 struct virtio_balloon *vb;
73 unsigned int len;
74
75 vb = virtqueue_get_buf(vq, &len);
76 if (vb)
77 complete(&vb->acked);
78}
79
80static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
81{
82 struct scatterlist sg;
83
84 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
85
86 init_completion(&vb->acked);
87
88 /* We should always be able to add one buffer to an empty queue. */
89 if (virtqueue_add_buf(vq, &sg, 1, 0, vb) < 0)
90 BUG();
91 virtqueue_kick(vq);
92
93 /* When host has read buffer, this completes via balloon_ack */
94 wait_for_completion(&vb->acked);
95}
96
97static void fill_balloon(struct virtio_balloon *vb, size_t num)
98{
99 /* We can only do one array worth at a time. */
100 num = min(num, ARRAY_SIZE(vb->pfns));
101
102 for (vb->num_pfns = 0; vb->num_pfns < num; vb->num_pfns++) {
103 struct page *page = alloc_page(GFP_HIGHUSER | __GFP_NORETRY |
104 __GFP_NOMEMALLOC | __GFP_NOWARN);
105 if (!page) {
106 if (printk_ratelimit())
107 dev_printk(KERN_INFO, &vb->vdev->dev,
108 "Out of puff! Can't get %zu pages\n",
109 num);
110 /* Sleep for at least 1/5 of a second before retry. */
111 msleep(200);
112 break;
113 }
114 vb->pfns[vb->num_pfns] = page_to_balloon_pfn(page);
115 totalram_pages--;
116 vb->num_pages++;
117 list_add(&page->lru, &vb->pages);
118 }
119
120 /* Didn't get any? Oh well. */
121 if (vb->num_pfns == 0)
122 return;
123
124 tell_host(vb, vb->inflate_vq);
125}
126
127static void release_pages_by_pfn(const u32 pfns[], unsigned int num)
128{
129 unsigned int i;
130
131 for (i = 0; i < num; i++) {
132 __free_page(pfn_to_page(pfns[i]));
133 totalram_pages++;
134 }
135}
136
137static void leak_balloon(struct virtio_balloon *vb, size_t num)
138{
139 struct page *page;
140
141 /* We can only do one array worth at a time. */
142 num = min(num, ARRAY_SIZE(vb->pfns));
143
144 for (vb->num_pfns = 0; vb->num_pfns < num; vb->num_pfns++) {
145 page = list_first_entry(&vb->pages, struct page, lru);
146 list_del(&page->lru);
147 vb->pfns[vb->num_pfns] = page_to_balloon_pfn(page);
148 vb->num_pages--;
149 }
150
151
152 /*
153 * Note that if
154 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
155 * is true, we *have* to do it in this order
156 */
157 tell_host(vb, vb->deflate_vq);
158 release_pages_by_pfn(vb->pfns, vb->num_pfns);
159}
160
161static inline void update_stat(struct virtio_balloon *vb, int idx,
162 u16 tag, u64 val)
163{
164 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
165 vb->stats[idx].tag = tag;
166 vb->stats[idx].val = val;
167}
168
169#define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
170
171static void update_balloon_stats(struct virtio_balloon *vb)
172{
173 unsigned long events[NR_VM_EVENT_ITEMS];
174 struct sysinfo i;
175 int idx = 0;
176
177 all_vm_events(events);
178 si_meminfo(&i);
179
180 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
181 pages_to_bytes(events[PSWPIN]));
182 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
183 pages_to_bytes(events[PSWPOUT]));
184 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
185 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
186 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
187 pages_to_bytes(i.freeram));
188 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
189 pages_to_bytes(i.totalram));
190}
191
192/*
193 * While most virtqueues communicate guest-initiated requests to the hypervisor,
194 * the stats queue operates in reverse. The driver initializes the virtqueue
195 * with a single buffer. From that point forward, all conversations consist of
196 * a hypervisor request (a call to this function) which directs us to refill
197 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
198 * we notify our kthread which does the actual work via stats_handle_request().
199 */
200static void stats_request(struct virtqueue *vq)
201{
202 struct virtio_balloon *vb;
203 unsigned int len;
204
205 vb = virtqueue_get_buf(vq, &len);
206 if (!vb)
207 return;
208 vb->need_stats_update = 1;
209 wake_up(&vb->config_change);
210}
211
212static void stats_handle_request(struct virtio_balloon *vb)
213{
214 struct virtqueue *vq;
215 struct scatterlist sg;
216
217 vb->need_stats_update = 0;
218 update_balloon_stats(vb);
219
220 vq = vb->stats_vq;
221 sg_init_one(&sg, vb->stats, sizeof(vb->stats));
222 if (virtqueue_add_buf(vq, &sg, 1, 0, vb) < 0)
223 BUG();
224 virtqueue_kick(vq);
225}
226
227static void virtballoon_changed(struct virtio_device *vdev)
228{
229 struct virtio_balloon *vb = vdev->priv;
230
231 wake_up(&vb->config_change);
232}
233
234static inline s64 towards_target(struct virtio_balloon *vb)
235{
236 u32 v;
237 vb->vdev->config->get(vb->vdev,
238 offsetof(struct virtio_balloon_config, num_pages),
239 &v, sizeof(v));
240 return (s64)v - vb->num_pages;
241}
242
243static void update_balloon_size(struct virtio_balloon *vb)
244{
245 __le32 actual = cpu_to_le32(vb->num_pages);
246
247 vb->vdev->config->set(vb->vdev,
248 offsetof(struct virtio_balloon_config, actual),
249 &actual, sizeof(actual));
250}
251
252static int balloon(void *_vballoon)
253{
254 struct virtio_balloon *vb = _vballoon;
255
256 set_freezable();
257 while (!kthread_should_stop()) {
258 s64 diff;
259
260 try_to_freeze();
261 wait_event_interruptible(vb->config_change,
262 (diff = towards_target(vb)) != 0
263 || vb->need_stats_update
264 || kthread_should_stop()
265 || freezing(current));
266 if (vb->need_stats_update)
267 stats_handle_request(vb);
268 if (diff > 0)
269 fill_balloon(vb, diff);
270 else if (diff < 0)
271 leak_balloon(vb, -diff);
272 update_balloon_size(vb);
273 }
274 return 0;
275}
276
277static int virtballoon_probe(struct virtio_device *vdev)
278{
279 struct virtio_balloon *vb;
280 struct virtqueue *vqs[3];
281 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
282 const char *names[] = { "inflate", "deflate", "stats" };
283 int err, nvqs;
284
285 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
286 if (!vb) {
287 err = -ENOMEM;
288 goto out;
289 }
290
291 INIT_LIST_HEAD(&vb->pages);
292 vb->num_pages = 0;
293 init_waitqueue_head(&vb->config_change);
294 vb->vdev = vdev;
295 vb->need_stats_update = 0;
296
297 /* We expect two virtqueues: inflate and deflate,
298 * and optionally stat. */
299 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
300 err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
301 if (err)
302 goto out_free_vb;
303
304 vb->inflate_vq = vqs[0];
305 vb->deflate_vq = vqs[1];
306 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
307 struct scatterlist sg;
308 vb->stats_vq = vqs[2];
309
310 /*
311 * Prime this virtqueue with one buffer so the hypervisor can
312 * use it to signal us later.
313 */
314 sg_init_one(&sg, vb->stats, sizeof vb->stats);
315 if (virtqueue_add_buf(vb->stats_vq, &sg, 1, 0, vb) < 0)
316 BUG();
317 virtqueue_kick(vb->stats_vq);
318 }
319
320 vb->thread = kthread_run(balloon, vb, "vballoon");
321 if (IS_ERR(vb->thread)) {
322 err = PTR_ERR(vb->thread);
323 goto out_del_vqs;
324 }
325
326 return 0;
327
328out_del_vqs:
329 vdev->config->del_vqs(vdev);
330out_free_vb:
331 kfree(vb);
332out:
333 return err;
334}
335
336static void __devexit virtballoon_remove(struct virtio_device *vdev)
337{
338 struct virtio_balloon *vb = vdev->priv;
339
340 kthread_stop(vb->thread);
341
342 /* There might be pages left in the balloon: free them. */
343 while (vb->num_pages)
344 leak_balloon(vb, vb->num_pages);
345
346 /* Now we reset the device so we can clean up the queues. */
347 vdev->config->reset(vdev);
348
349 vdev->config->del_vqs(vdev);
350 kfree(vb);
351}
352
353static unsigned int features[] = {
354 VIRTIO_BALLOON_F_MUST_TELL_HOST,
355 VIRTIO_BALLOON_F_STATS_VQ,
356};
357
358static struct virtio_driver virtio_balloon_driver = {
359 .feature_table = features,
360 .feature_table_size = ARRAY_SIZE(features),
361 .driver.name = KBUILD_MODNAME,
362 .driver.owner = THIS_MODULE,
363 .id_table = id_table,
364 .probe = virtballoon_probe,
365 .remove = __devexit_p(virtballoon_remove),
366 .config_changed = virtballoon_changed,
367};
368
369static int __init init(void)
370{
371 return register_virtio_driver(&virtio_balloon_driver);
372}
373
374static void __exit fini(void)
375{
376 unregister_virtio_driver(&virtio_balloon_driver);
377}
378module_init(init);
379module_exit(fini);
380
381MODULE_DEVICE_TABLE(virtio, id_table);
382MODULE_DESCRIPTION("Virtio balloon driver");
383MODULE_LICENSE("GPL");
1/*
2 * Virtio balloon implementation, inspired by Dor Laor and Marcelo
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 */
21
22#include <linux/virtio.h>
23#include <linux/virtio_balloon.h>
24#include <linux/swap.h>
25#include <linux/workqueue.h>
26#include <linux/delay.h>
27#include <linux/slab.h>
28#include <linux/module.h>
29#include <linux/balloon_compaction.h>
30#include <linux/oom.h>
31#include <linux/wait.h>
32#include <linux/mm.h>
33#include <linux/mount.h>
34#include <linux/magic.h>
35
36/*
37 * Balloon device works in 4K page units. So each page is pointed to by
38 * multiple balloon pages. All memory counters in this driver are in balloon
39 * page units.
40 */
41#define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
42#define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
43#define OOM_VBALLOON_DEFAULT_PAGES 256
44#define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
45
46static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES;
47module_param(oom_pages, int, S_IRUSR | S_IWUSR);
48MODULE_PARM_DESC(oom_pages, "pages to free on OOM");
49
50#ifdef CONFIG_BALLOON_COMPACTION
51static struct vfsmount *balloon_mnt;
52#endif
53
54struct virtio_balloon {
55 struct virtio_device *vdev;
56 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
57
58 /* The balloon servicing is delegated to a freezable workqueue. */
59 struct work_struct update_balloon_stats_work;
60 struct work_struct update_balloon_size_work;
61
62 /* Prevent updating balloon when it is being canceled. */
63 spinlock_t stop_update_lock;
64 bool stop_update;
65
66 /* Waiting for host to ack the pages we released. */
67 wait_queue_head_t acked;
68
69 /* Number of balloon pages we've told the Host we're not using. */
70 unsigned int num_pages;
71 /*
72 * The pages we've told the Host we're not using are enqueued
73 * at vb_dev_info->pages list.
74 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
75 * to num_pages above.
76 */
77 struct balloon_dev_info vb_dev_info;
78
79 /* Synchronize access/update to this struct virtio_balloon elements */
80 struct mutex balloon_lock;
81
82 /* The array of pfns we tell the Host about. */
83 unsigned int num_pfns;
84 __virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
85
86 /* Memory statistics */
87 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
88
89 /* To register callback in oom notifier call chain */
90 struct notifier_block nb;
91};
92
93static struct virtio_device_id id_table[] = {
94 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
95 { 0 },
96};
97
98static u32 page_to_balloon_pfn(struct page *page)
99{
100 unsigned long pfn = page_to_pfn(page);
101
102 BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
103 /* Convert pfn from Linux page size to balloon page size. */
104 return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
105}
106
107static void balloon_ack(struct virtqueue *vq)
108{
109 struct virtio_balloon *vb = vq->vdev->priv;
110
111 wake_up(&vb->acked);
112}
113
114static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
115{
116 struct scatterlist sg;
117 unsigned int len;
118
119 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
120
121 /* We should always be able to add one buffer to an empty queue. */
122 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
123 virtqueue_kick(vq);
124
125 /* When host has read buffer, this completes via balloon_ack */
126 wait_event(vb->acked, virtqueue_get_buf(vq, &len));
127
128}
129
130static void set_page_pfns(struct virtio_balloon *vb,
131 __virtio32 pfns[], struct page *page)
132{
133 unsigned int i;
134
135 /*
136 * Set balloon pfns pointing at this page.
137 * Note that the first pfn points at start of the page.
138 */
139 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
140 pfns[i] = cpu_to_virtio32(vb->vdev,
141 page_to_balloon_pfn(page) + i);
142}
143
144static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
145{
146 unsigned num_allocated_pages;
147 unsigned num_pfns;
148 struct page *page;
149 LIST_HEAD(pages);
150
151 /* We can only do one array worth at a time. */
152 num = min(num, ARRAY_SIZE(vb->pfns));
153
154 for (num_pfns = 0; num_pfns < num;
155 num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
156 struct page *page = balloon_page_alloc();
157
158 if (!page) {
159 dev_info_ratelimited(&vb->vdev->dev,
160 "Out of puff! Can't get %u pages\n",
161 VIRTIO_BALLOON_PAGES_PER_PAGE);
162 /* Sleep for at least 1/5 of a second before retry. */
163 msleep(200);
164 break;
165 }
166
167 balloon_page_push(&pages, page);
168 }
169
170 mutex_lock(&vb->balloon_lock);
171
172 vb->num_pfns = 0;
173
174 while ((page = balloon_page_pop(&pages))) {
175 balloon_page_enqueue(&vb->vb_dev_info, page);
176
177 set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
178 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
179 if (!virtio_has_feature(vb->vdev,
180 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
181 adjust_managed_page_count(page, -1);
182 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
183 }
184
185 num_allocated_pages = vb->num_pfns;
186 /* Did we get any? */
187 if (vb->num_pfns != 0)
188 tell_host(vb, vb->inflate_vq);
189 mutex_unlock(&vb->balloon_lock);
190
191 return num_allocated_pages;
192}
193
194static void release_pages_balloon(struct virtio_balloon *vb,
195 struct list_head *pages)
196{
197 struct page *page, *next;
198
199 list_for_each_entry_safe(page, next, pages, lru) {
200 if (!virtio_has_feature(vb->vdev,
201 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
202 adjust_managed_page_count(page, 1);
203 list_del(&page->lru);
204 put_page(page); /* balloon reference */
205 }
206}
207
208static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
209{
210 unsigned num_freed_pages;
211 struct page *page;
212 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
213 LIST_HEAD(pages);
214
215 /* We can only do one array worth at a time. */
216 num = min(num, ARRAY_SIZE(vb->pfns));
217
218 mutex_lock(&vb->balloon_lock);
219 /* We can't release more pages than taken */
220 num = min(num, (size_t)vb->num_pages);
221 for (vb->num_pfns = 0; vb->num_pfns < num;
222 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
223 page = balloon_page_dequeue(vb_dev_info);
224 if (!page)
225 break;
226 set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
227 list_add(&page->lru, &pages);
228 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
229 }
230
231 num_freed_pages = vb->num_pfns;
232 /*
233 * Note that if
234 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
235 * is true, we *have* to do it in this order
236 */
237 if (vb->num_pfns != 0)
238 tell_host(vb, vb->deflate_vq);
239 release_pages_balloon(vb, &pages);
240 mutex_unlock(&vb->balloon_lock);
241 return num_freed_pages;
242}
243
244static inline void update_stat(struct virtio_balloon *vb, int idx,
245 u16 tag, u64 val)
246{
247 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
248 vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
249 vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
250}
251
252#define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
253
254static unsigned int update_balloon_stats(struct virtio_balloon *vb)
255{
256 unsigned long events[NR_VM_EVENT_ITEMS];
257 struct sysinfo i;
258 unsigned int idx = 0;
259 long available;
260 unsigned long caches;
261
262 all_vm_events(events);
263 si_meminfo(&i);
264
265 available = si_mem_available();
266 caches = global_node_page_state(NR_FILE_PAGES);
267
268#ifdef CONFIG_VM_EVENT_COUNTERS
269 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
270 pages_to_bytes(events[PSWPIN]));
271 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
272 pages_to_bytes(events[PSWPOUT]));
273 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
274 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
275#ifdef CONFIG_HUGETLB_PAGE
276 update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC,
277 events[HTLB_BUDDY_PGALLOC]);
278 update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL,
279 events[HTLB_BUDDY_PGALLOC_FAIL]);
280#endif
281#endif
282 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
283 pages_to_bytes(i.freeram));
284 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
285 pages_to_bytes(i.totalram));
286 update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
287 pages_to_bytes(available));
288 update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES,
289 pages_to_bytes(caches));
290
291 return idx;
292}
293
294/*
295 * While most virtqueues communicate guest-initiated requests to the hypervisor,
296 * the stats queue operates in reverse. The driver initializes the virtqueue
297 * with a single buffer. From that point forward, all conversations consist of
298 * a hypervisor request (a call to this function) which directs us to refill
299 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
300 * we delegate the job to a freezable workqueue that will do the actual work via
301 * stats_handle_request().
302 */
303static void stats_request(struct virtqueue *vq)
304{
305 struct virtio_balloon *vb = vq->vdev->priv;
306
307 spin_lock(&vb->stop_update_lock);
308 if (!vb->stop_update)
309 queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
310 spin_unlock(&vb->stop_update_lock);
311}
312
313static void stats_handle_request(struct virtio_balloon *vb)
314{
315 struct virtqueue *vq;
316 struct scatterlist sg;
317 unsigned int len, num_stats;
318
319 num_stats = update_balloon_stats(vb);
320
321 vq = vb->stats_vq;
322 if (!virtqueue_get_buf(vq, &len))
323 return;
324 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
325 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
326 virtqueue_kick(vq);
327}
328
329static void virtballoon_changed(struct virtio_device *vdev)
330{
331 struct virtio_balloon *vb = vdev->priv;
332 unsigned long flags;
333
334 spin_lock_irqsave(&vb->stop_update_lock, flags);
335 if (!vb->stop_update)
336 queue_work(system_freezable_wq, &vb->update_balloon_size_work);
337 spin_unlock_irqrestore(&vb->stop_update_lock, flags);
338}
339
340static inline s64 towards_target(struct virtio_balloon *vb)
341{
342 s64 target;
343 u32 num_pages;
344
345 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
346 &num_pages);
347
348 /* Legacy balloon config space is LE, unlike all other devices. */
349 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
350 num_pages = le32_to_cpu((__force __le32)num_pages);
351
352 target = num_pages;
353 return target - vb->num_pages;
354}
355
356static void update_balloon_size(struct virtio_balloon *vb)
357{
358 u32 actual = vb->num_pages;
359
360 /* Legacy balloon config space is LE, unlike all other devices. */
361 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
362 actual = (__force u32)cpu_to_le32(actual);
363
364 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
365 &actual);
366}
367
368/*
369 * virtballoon_oom_notify - release pages when system is under severe
370 * memory pressure (called from out_of_memory())
371 * @self : notifier block struct
372 * @dummy: not used
373 * @parm : returned - number of freed pages
374 *
375 * The balancing of memory by use of the virtio balloon should not cause
376 * the termination of processes while there are pages in the balloon.
377 * If virtio balloon manages to release some memory, it will make the
378 * system return and retry the allocation that forced the OOM killer
379 * to run.
380 */
381static int virtballoon_oom_notify(struct notifier_block *self,
382 unsigned long dummy, void *parm)
383{
384 struct virtio_balloon *vb;
385 unsigned long *freed;
386 unsigned num_freed_pages;
387
388 vb = container_of(self, struct virtio_balloon, nb);
389 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
390 return NOTIFY_OK;
391
392 freed = parm;
393 num_freed_pages = leak_balloon(vb, oom_pages);
394 update_balloon_size(vb);
395 *freed += num_freed_pages;
396
397 return NOTIFY_OK;
398}
399
400static void update_balloon_stats_func(struct work_struct *work)
401{
402 struct virtio_balloon *vb;
403
404 vb = container_of(work, struct virtio_balloon,
405 update_balloon_stats_work);
406 stats_handle_request(vb);
407}
408
409static void update_balloon_size_func(struct work_struct *work)
410{
411 struct virtio_balloon *vb;
412 s64 diff;
413
414 vb = container_of(work, struct virtio_balloon,
415 update_balloon_size_work);
416 diff = towards_target(vb);
417
418 if (diff > 0)
419 diff -= fill_balloon(vb, diff);
420 else if (diff < 0)
421 diff += leak_balloon(vb, -diff);
422 update_balloon_size(vb);
423
424 if (diff)
425 queue_work(system_freezable_wq, work);
426}
427
428static int init_vqs(struct virtio_balloon *vb)
429{
430 struct virtqueue *vqs[3];
431 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
432 static const char * const names[] = { "inflate", "deflate", "stats" };
433 int err, nvqs;
434
435 /*
436 * We expect two virtqueues: inflate and deflate, and
437 * optionally stat.
438 */
439 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
440 err = virtio_find_vqs(vb->vdev, nvqs, vqs, callbacks, names, NULL);
441 if (err)
442 return err;
443
444 vb->inflate_vq = vqs[0];
445 vb->deflate_vq = vqs[1];
446 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
447 struct scatterlist sg;
448 unsigned int num_stats;
449 vb->stats_vq = vqs[2];
450
451 /*
452 * Prime this virtqueue with one buffer so the hypervisor can
453 * use it to signal us later (it can't be broken yet!).
454 */
455 num_stats = update_balloon_stats(vb);
456
457 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
458 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
459 < 0)
460 BUG();
461 virtqueue_kick(vb->stats_vq);
462 }
463 return 0;
464}
465
466#ifdef CONFIG_BALLOON_COMPACTION
467/*
468 * virtballoon_migratepage - perform the balloon page migration on behalf of
469 * a compation thread. (called under page lock)
470 * @vb_dev_info: the balloon device
471 * @newpage: page that will replace the isolated page after migration finishes.
472 * @page : the isolated (old) page that is about to be migrated to newpage.
473 * @mode : compaction mode -- not used for balloon page migration.
474 *
475 * After a ballooned page gets isolated by compaction procedures, this is the
476 * function that performs the page migration on behalf of a compaction thread
477 * The page migration for virtio balloon is done in a simple swap fashion which
478 * follows these two macro steps:
479 * 1) insert newpage into vb->pages list and update the host about it;
480 * 2) update the host about the old page removed from vb->pages list;
481 *
482 * This function preforms the balloon page migration task.
483 * Called through balloon_mapping->a_ops->migratepage
484 */
485static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
486 struct page *newpage, struct page *page, enum migrate_mode mode)
487{
488 struct virtio_balloon *vb = container_of(vb_dev_info,
489 struct virtio_balloon, vb_dev_info);
490 unsigned long flags;
491
492 /*
493 * In order to avoid lock contention while migrating pages concurrently
494 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
495 * this turn, as it is easier to retry the page migration later.
496 * This also prevents fill_balloon() getting stuck into a mutex
497 * recursion in the case it ends up triggering memory compaction
498 * while it is attempting to inflate the ballon.
499 */
500 if (!mutex_trylock(&vb->balloon_lock))
501 return -EAGAIN;
502
503 get_page(newpage); /* balloon reference */
504
505 /* balloon's page migration 1st step -- inflate "newpage" */
506 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
507 balloon_page_insert(vb_dev_info, newpage);
508 vb_dev_info->isolated_pages--;
509 __count_vm_event(BALLOON_MIGRATE);
510 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
511 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
512 set_page_pfns(vb, vb->pfns, newpage);
513 tell_host(vb, vb->inflate_vq);
514
515 /* balloon's page migration 2nd step -- deflate "page" */
516 balloon_page_delete(page);
517 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
518 set_page_pfns(vb, vb->pfns, page);
519 tell_host(vb, vb->deflate_vq);
520
521 mutex_unlock(&vb->balloon_lock);
522
523 put_page(page); /* balloon reference */
524
525 return MIGRATEPAGE_SUCCESS;
526}
527
528static struct dentry *balloon_mount(struct file_system_type *fs_type,
529 int flags, const char *dev_name, void *data)
530{
531 static const struct dentry_operations ops = {
532 .d_dname = simple_dname,
533 };
534
535 return mount_pseudo(fs_type, "balloon-kvm:", NULL, &ops,
536 BALLOON_KVM_MAGIC);
537}
538
539static struct file_system_type balloon_fs = {
540 .name = "balloon-kvm",
541 .mount = balloon_mount,
542 .kill_sb = kill_anon_super,
543};
544
545#endif /* CONFIG_BALLOON_COMPACTION */
546
547static int virtballoon_probe(struct virtio_device *vdev)
548{
549 struct virtio_balloon *vb;
550 int err;
551
552 if (!vdev->config->get) {
553 dev_err(&vdev->dev, "%s failure: config access disabled\n",
554 __func__);
555 return -EINVAL;
556 }
557
558 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
559 if (!vb) {
560 err = -ENOMEM;
561 goto out;
562 }
563
564 INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
565 INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
566 spin_lock_init(&vb->stop_update_lock);
567 vb->stop_update = false;
568 vb->num_pages = 0;
569 mutex_init(&vb->balloon_lock);
570 init_waitqueue_head(&vb->acked);
571 vb->vdev = vdev;
572
573 balloon_devinfo_init(&vb->vb_dev_info);
574
575 err = init_vqs(vb);
576 if (err)
577 goto out_free_vb;
578
579 vb->nb.notifier_call = virtballoon_oom_notify;
580 vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
581 err = register_oom_notifier(&vb->nb);
582 if (err < 0)
583 goto out_del_vqs;
584
585#ifdef CONFIG_BALLOON_COMPACTION
586 balloon_mnt = kern_mount(&balloon_fs);
587 if (IS_ERR(balloon_mnt)) {
588 err = PTR_ERR(balloon_mnt);
589 unregister_oom_notifier(&vb->nb);
590 goto out_del_vqs;
591 }
592
593 vb->vb_dev_info.migratepage = virtballoon_migratepage;
594 vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb);
595 if (IS_ERR(vb->vb_dev_info.inode)) {
596 err = PTR_ERR(vb->vb_dev_info.inode);
597 kern_unmount(balloon_mnt);
598 unregister_oom_notifier(&vb->nb);
599 vb->vb_dev_info.inode = NULL;
600 goto out_del_vqs;
601 }
602 vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops;
603#endif
604
605 virtio_device_ready(vdev);
606
607 if (towards_target(vb))
608 virtballoon_changed(vdev);
609 return 0;
610
611out_del_vqs:
612 vdev->config->del_vqs(vdev);
613out_free_vb:
614 kfree(vb);
615out:
616 return err;
617}
618
619static void remove_common(struct virtio_balloon *vb)
620{
621 /* There might be pages left in the balloon: free them. */
622 while (vb->num_pages)
623 leak_balloon(vb, vb->num_pages);
624 update_balloon_size(vb);
625
626 /* Now we reset the device so we can clean up the queues. */
627 vb->vdev->config->reset(vb->vdev);
628
629 vb->vdev->config->del_vqs(vb->vdev);
630}
631
632static void virtballoon_remove(struct virtio_device *vdev)
633{
634 struct virtio_balloon *vb = vdev->priv;
635
636 unregister_oom_notifier(&vb->nb);
637
638 spin_lock_irq(&vb->stop_update_lock);
639 vb->stop_update = true;
640 spin_unlock_irq(&vb->stop_update_lock);
641 cancel_work_sync(&vb->update_balloon_size_work);
642 cancel_work_sync(&vb->update_balloon_stats_work);
643
644 remove_common(vb);
645#ifdef CONFIG_BALLOON_COMPACTION
646 if (vb->vb_dev_info.inode)
647 iput(vb->vb_dev_info.inode);
648
649 kern_unmount(balloon_mnt);
650#endif
651 kfree(vb);
652}
653
654#ifdef CONFIG_PM_SLEEP
655static int virtballoon_freeze(struct virtio_device *vdev)
656{
657 struct virtio_balloon *vb = vdev->priv;
658
659 /*
660 * The workqueue is already frozen by the PM core before this
661 * function is called.
662 */
663 remove_common(vb);
664 return 0;
665}
666
667static int virtballoon_restore(struct virtio_device *vdev)
668{
669 struct virtio_balloon *vb = vdev->priv;
670 int ret;
671
672 ret = init_vqs(vdev->priv);
673 if (ret)
674 return ret;
675
676 virtio_device_ready(vdev);
677
678 if (towards_target(vb))
679 virtballoon_changed(vdev);
680 update_balloon_size(vb);
681 return 0;
682}
683#endif
684
685static int virtballoon_validate(struct virtio_device *vdev)
686{
687 __virtio_clear_bit(vdev, VIRTIO_F_IOMMU_PLATFORM);
688 return 0;
689}
690
691static unsigned int features[] = {
692 VIRTIO_BALLOON_F_MUST_TELL_HOST,
693 VIRTIO_BALLOON_F_STATS_VQ,
694 VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
695};
696
697static struct virtio_driver virtio_balloon_driver = {
698 .feature_table = features,
699 .feature_table_size = ARRAY_SIZE(features),
700 .driver.name = KBUILD_MODNAME,
701 .driver.owner = THIS_MODULE,
702 .id_table = id_table,
703 .validate = virtballoon_validate,
704 .probe = virtballoon_probe,
705 .remove = virtballoon_remove,
706 .config_changed = virtballoon_changed,
707#ifdef CONFIG_PM_SLEEP
708 .freeze = virtballoon_freeze,
709 .restore = virtballoon_restore,
710#endif
711};
712
713module_virtio_driver(virtio_balloon_driver);
714MODULE_DEVICE_TABLE(virtio, id_table);
715MODULE_DESCRIPTION("Virtio balloon driver");
716MODULE_LICENSE("GPL");