2 * Virtio balloon implementation, inspired by Dor Laor and Marcelo
3 * Tosatti's implementations.
5 * Copyright 2008 Rusty Russell IBM Corporation
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.
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.
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
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>
33 #include <linux/mount.h>
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
40 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
41 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
42 #define OOM_VBALLOON_DEFAULT_PAGES 256
43 #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
45 static int oom_pages
= OOM_VBALLOON_DEFAULT_PAGES
;
46 module_param(oom_pages
, int, S_IRUSR
| S_IWUSR
);
47 MODULE_PARM_DESC(oom_pages
, "pages to free on OOM");
49 #ifdef CONFIG_BALLOON_COMPACTION
50 static struct vfsmount
*balloon_mnt
;
53 struct virtio_balloon
{
54 struct virtio_device
*vdev
;
55 struct virtqueue
*inflate_vq
, *deflate_vq
, *stats_vq
;
57 /* The balloon servicing is delegated to a freezable workqueue. */
58 struct work_struct update_balloon_stats_work
;
59 struct work_struct update_balloon_size_work
;
61 /* Prevent updating balloon when it is being canceled. */
62 spinlock_t stop_update_lock
;
65 /* Waiting for host to ack the pages we released. */
66 wait_queue_head_t acked
;
68 /* Number of balloon pages we've told the Host we're not using. */
69 unsigned int num_pages
;
71 * The pages we've told the Host we're not using are enqueued
72 * at vb_dev_info->pages list.
73 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
76 struct balloon_dev_info vb_dev_info
;
78 /* Synchronize access/update to this struct virtio_balloon elements */
79 struct mutex balloon_lock
;
81 /* The array of pfns we tell the Host about. */
82 unsigned int num_pfns
;
83 __virtio32 pfns
[VIRTIO_BALLOON_ARRAY_PFNS_MAX
];
85 /* Memory statistics */
86 struct virtio_balloon_stat stats
[VIRTIO_BALLOON_S_NR
];
88 /* To register callback in oom notifier call chain */
89 struct notifier_block nb
;
92 static struct virtio_device_id id_table
[] = {
93 { VIRTIO_ID_BALLOON
, VIRTIO_DEV_ANY_ID
},
97 static u32
page_to_balloon_pfn(struct page
*page
)
99 unsigned long pfn
= page_to_pfn(page
);
101 BUILD_BUG_ON(PAGE_SHIFT
< VIRTIO_BALLOON_PFN_SHIFT
);
102 /* Convert pfn from Linux page size to balloon page size. */
103 return pfn
* VIRTIO_BALLOON_PAGES_PER_PAGE
;
106 static struct page
*balloon_pfn_to_page(u32 pfn
)
108 BUG_ON(pfn
% VIRTIO_BALLOON_PAGES_PER_PAGE
);
109 return pfn_to_page(pfn
/ VIRTIO_BALLOON_PAGES_PER_PAGE
);
112 static void balloon_ack(struct virtqueue
*vq
)
114 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
119 static void tell_host(struct virtio_balloon
*vb
, struct virtqueue
*vq
)
121 struct scatterlist sg
;
124 sg_init_one(&sg
, vb
->pfns
, sizeof(vb
->pfns
[0]) * vb
->num_pfns
);
126 /* We should always be able to add one buffer to an empty queue. */
127 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
130 /* When host has read buffer, this completes via balloon_ack */
131 wait_event(vb
->acked
, virtqueue_get_buf(vq
, &len
));
135 static void set_page_pfns(struct virtio_balloon
*vb
,
136 __virtio32 pfns
[], struct page
*page
)
140 /* Set balloon pfns pointing at this page.
141 * Note that the first pfn points at start of the page. */
142 for (i
= 0; i
< VIRTIO_BALLOON_PAGES_PER_PAGE
; i
++)
143 pfns
[i
] = cpu_to_virtio32(vb
->vdev
,
144 page_to_balloon_pfn(page
) + i
);
147 static unsigned fill_balloon(struct virtio_balloon
*vb
, size_t num
)
149 struct balloon_dev_info
*vb_dev_info
= &vb
->vb_dev_info
;
150 unsigned num_allocated_pages
;
152 /* We can only do one array worth at a time. */
153 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
155 mutex_lock(&vb
->balloon_lock
);
156 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
157 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
158 struct page
*page
= balloon_page_enqueue(vb_dev_info
);
161 dev_info_ratelimited(&vb
->vdev
->dev
,
162 "Out of puff! Can't get %u pages\n",
163 VIRTIO_BALLOON_PAGES_PER_PAGE
);
164 /* Sleep for at least 1/5 of a second before retry. */
168 set_page_pfns(vb
, vb
->pfns
+ vb
->num_pfns
, page
);
169 vb
->num_pages
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
170 if (!virtio_has_feature(vb
->vdev
,
171 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
172 adjust_managed_page_count(page
, -1);
175 num_allocated_pages
= vb
->num_pfns
;
176 /* Did we get any? */
177 if (vb
->num_pfns
!= 0)
178 tell_host(vb
, vb
->inflate_vq
);
179 mutex_unlock(&vb
->balloon_lock
);
181 return num_allocated_pages
;
184 static void release_pages_balloon(struct virtio_balloon
*vb
)
189 /* Find pfns pointing at start of each page, get pages and free them. */
190 for (i
= 0; i
< vb
->num_pfns
; i
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
191 page
= balloon_pfn_to_page(virtio32_to_cpu(vb
->vdev
,
193 if (!virtio_has_feature(vb
->vdev
,
194 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
195 adjust_managed_page_count(page
, 1);
196 put_page(page
); /* balloon reference */
200 static unsigned leak_balloon(struct virtio_balloon
*vb
, size_t num
)
202 unsigned num_freed_pages
;
204 struct balloon_dev_info
*vb_dev_info
= &vb
->vb_dev_info
;
206 /* We can only do one array worth at a time. */
207 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
209 mutex_lock(&vb
->balloon_lock
);
210 /* We can't release more pages than taken */
211 num
= min(num
, (size_t)vb
->num_pages
);
212 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
213 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
214 page
= balloon_page_dequeue(vb_dev_info
);
217 set_page_pfns(vb
, vb
->pfns
+ vb
->num_pfns
, page
);
218 vb
->num_pages
-= VIRTIO_BALLOON_PAGES_PER_PAGE
;
221 num_freed_pages
= vb
->num_pfns
;
224 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
225 * is true, we *have* to do it in this order
227 if (vb
->num_pfns
!= 0)
228 tell_host(vb
, vb
->deflate_vq
);
229 release_pages_balloon(vb
);
230 mutex_unlock(&vb
->balloon_lock
);
231 return num_freed_pages
;
234 static inline void update_stat(struct virtio_balloon
*vb
, int idx
,
237 BUG_ON(idx
>= VIRTIO_BALLOON_S_NR
);
238 vb
->stats
[idx
].tag
= cpu_to_virtio16(vb
->vdev
, tag
);
239 vb
->stats
[idx
].val
= cpu_to_virtio64(vb
->vdev
, val
);
242 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
244 static void update_balloon_stats(struct virtio_balloon
*vb
)
246 unsigned long events
[NR_VM_EVENT_ITEMS
];
251 all_vm_events(events
);
254 available
= si_mem_available();
256 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_IN
,
257 pages_to_bytes(events
[PSWPIN
]));
258 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_OUT
,
259 pages_to_bytes(events
[PSWPOUT
]));
260 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MAJFLT
, events
[PGMAJFAULT
]);
261 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MINFLT
, events
[PGFAULT
]);
262 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMFREE
,
263 pages_to_bytes(i
.freeram
));
264 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMTOT
,
265 pages_to_bytes(i
.totalram
));
266 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_AVAIL
,
267 pages_to_bytes(available
));
271 * While most virtqueues communicate guest-initiated requests to the hypervisor,
272 * the stats queue operates in reverse. The driver initializes the virtqueue
273 * with a single buffer. From that point forward, all conversations consist of
274 * a hypervisor request (a call to this function) which directs us to refill
275 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
276 * we delegate the job to a freezable workqueue that will do the actual work via
277 * stats_handle_request().
279 static void stats_request(struct virtqueue
*vq
)
281 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
283 spin_lock(&vb
->stop_update_lock
);
284 if (!vb
->stop_update
)
285 queue_work(system_freezable_wq
, &vb
->update_balloon_stats_work
);
286 spin_unlock(&vb
->stop_update_lock
);
289 static void stats_handle_request(struct virtio_balloon
*vb
)
291 struct virtqueue
*vq
;
292 struct scatterlist sg
;
295 update_balloon_stats(vb
);
298 if (!virtqueue_get_buf(vq
, &len
))
300 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
));
301 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
305 static void virtballoon_changed(struct virtio_device
*vdev
)
307 struct virtio_balloon
*vb
= vdev
->priv
;
310 spin_lock_irqsave(&vb
->stop_update_lock
, flags
);
311 if (!vb
->stop_update
)
312 queue_work(system_freezable_wq
, &vb
->update_balloon_size_work
);
313 spin_unlock_irqrestore(&vb
->stop_update_lock
, flags
);
316 static inline s64
towards_target(struct virtio_balloon
*vb
)
321 virtio_cread(vb
->vdev
, struct virtio_balloon_config
, num_pages
,
324 /* Legacy balloon config space is LE, unlike all other devices. */
325 if (!virtio_has_feature(vb
->vdev
, VIRTIO_F_VERSION_1
))
326 num_pages
= le32_to_cpu((__force __le32
)num_pages
);
329 return target
- vb
->num_pages
;
332 static void update_balloon_size(struct virtio_balloon
*vb
)
334 u32 actual
= vb
->num_pages
;
336 /* Legacy balloon config space is LE, unlike all other devices. */
337 if (!virtio_has_feature(vb
->vdev
, VIRTIO_F_VERSION_1
))
338 actual
= (__force u32
)cpu_to_le32(actual
);
340 virtio_cwrite(vb
->vdev
, struct virtio_balloon_config
, actual
,
345 * virtballoon_oom_notify - release pages when system is under severe
346 * memory pressure (called from out_of_memory())
347 * @self : notifier block struct
349 * @parm : returned - number of freed pages
351 * The balancing of memory by use of the virtio balloon should not cause
352 * the termination of processes while there are pages in the balloon.
353 * If virtio balloon manages to release some memory, it will make the
354 * system return and retry the allocation that forced the OOM killer
357 static int virtballoon_oom_notify(struct notifier_block
*self
,
358 unsigned long dummy
, void *parm
)
360 struct virtio_balloon
*vb
;
361 unsigned long *freed
;
362 unsigned num_freed_pages
;
364 vb
= container_of(self
, struct virtio_balloon
, nb
);
365 if (!virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
369 num_freed_pages
= leak_balloon(vb
, oom_pages
);
370 update_balloon_size(vb
);
371 *freed
+= num_freed_pages
;
376 static void update_balloon_stats_func(struct work_struct
*work
)
378 struct virtio_balloon
*vb
;
380 vb
= container_of(work
, struct virtio_balloon
,
381 update_balloon_stats_work
);
382 stats_handle_request(vb
);
385 static void update_balloon_size_func(struct work_struct
*work
)
387 struct virtio_balloon
*vb
;
390 vb
= container_of(work
, struct virtio_balloon
,
391 update_balloon_size_work
);
392 diff
= towards_target(vb
);
395 diff
-= fill_balloon(vb
, diff
);
397 diff
+= leak_balloon(vb
, -diff
);
398 update_balloon_size(vb
);
401 queue_work(system_freezable_wq
, work
);
404 static int init_vqs(struct virtio_balloon
*vb
)
406 struct virtqueue
*vqs
[3];
407 vq_callback_t
*callbacks
[] = { balloon_ack
, balloon_ack
, stats_request
};
408 static const char * const names
[] = { "inflate", "deflate", "stats" };
412 * We expect two virtqueues: inflate and deflate, and
415 nvqs
= virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
) ? 3 : 2;
416 err
= vb
->vdev
->config
->find_vqs(vb
->vdev
, nvqs
, vqs
, callbacks
, names
);
420 vb
->inflate_vq
= vqs
[0];
421 vb
->deflate_vq
= vqs
[1];
422 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
423 struct scatterlist sg
;
424 vb
->stats_vq
= vqs
[2];
427 * Prime this virtqueue with one buffer so the hypervisor can
428 * use it to signal us later (it can't be broken yet!).
430 sg_init_one(&sg
, vb
->stats
, sizeof vb
->stats
);
431 if (virtqueue_add_outbuf(vb
->stats_vq
, &sg
, 1, vb
, GFP_KERNEL
)
434 virtqueue_kick(vb
->stats_vq
);
439 #ifdef CONFIG_BALLOON_COMPACTION
441 * virtballoon_migratepage - perform the balloon page migration on behalf of
442 * a compation thread. (called under page lock)
443 * @vb_dev_info: the balloon device
444 * @newpage: page that will replace the isolated page after migration finishes.
445 * @page : the isolated (old) page that is about to be migrated to newpage.
446 * @mode : compaction mode -- not used for balloon page migration.
448 * After a ballooned page gets isolated by compaction procedures, this is the
449 * function that performs the page migration on behalf of a compaction thread
450 * The page migration for virtio balloon is done in a simple swap fashion which
451 * follows these two macro steps:
452 * 1) insert newpage into vb->pages list and update the host about it;
453 * 2) update the host about the old page removed from vb->pages list;
455 * This function preforms the balloon page migration task.
456 * Called through balloon_mapping->a_ops->migratepage
458 static int virtballoon_migratepage(struct balloon_dev_info
*vb_dev_info
,
459 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
461 struct virtio_balloon
*vb
= container_of(vb_dev_info
,
462 struct virtio_balloon
, vb_dev_info
);
466 * In order to avoid lock contention while migrating pages concurrently
467 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
468 * this turn, as it is easier to retry the page migration later.
469 * This also prevents fill_balloon() getting stuck into a mutex
470 * recursion in the case it ends up triggering memory compaction
471 * while it is attempting to inflate the ballon.
473 if (!mutex_trylock(&vb
->balloon_lock
))
476 get_page(newpage
); /* balloon reference */
478 /* balloon's page migration 1st step -- inflate "newpage" */
479 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
480 balloon_page_insert(vb_dev_info
, newpage
);
481 vb_dev_info
->isolated_pages
--;
482 __count_vm_event(BALLOON_MIGRATE
);
483 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
484 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
485 set_page_pfns(vb
, vb
->pfns
, newpage
);
486 tell_host(vb
, vb
->inflate_vq
);
488 /* balloon's page migration 2nd step -- deflate "page" */
489 balloon_page_delete(page
);
490 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
491 set_page_pfns(vb
, vb
->pfns
, page
);
492 tell_host(vb
, vb
->deflate_vq
);
494 mutex_unlock(&vb
->balloon_lock
);
496 put_page(page
); /* balloon reference */
498 return MIGRATEPAGE_SUCCESS
;
501 static struct dentry
*balloon_mount(struct file_system_type
*fs_type
,
502 int flags
, const char *dev_name
, void *data
)
504 static const struct dentry_operations ops
= {
505 .d_dname
= simple_dname
,
508 return mount_pseudo(fs_type
, "balloon-kvm:", NULL
, &ops
,
512 static struct file_system_type balloon_fs
= {
513 .name
= "balloon-kvm",
514 .mount
= balloon_mount
,
515 .kill_sb
= kill_anon_super
,
518 #endif /* CONFIG_BALLOON_COMPACTION */
520 static int virtballoon_probe(struct virtio_device
*vdev
)
522 struct virtio_balloon
*vb
;
525 if (!vdev
->config
->get
) {
526 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
531 vdev
->priv
= vb
= kmalloc(sizeof(*vb
), GFP_KERNEL
);
537 INIT_WORK(&vb
->update_balloon_stats_work
, update_balloon_stats_func
);
538 INIT_WORK(&vb
->update_balloon_size_work
, update_balloon_size_func
);
539 spin_lock_init(&vb
->stop_update_lock
);
540 vb
->stop_update
= false;
542 mutex_init(&vb
->balloon_lock
);
543 init_waitqueue_head(&vb
->acked
);
546 balloon_devinfo_init(&vb
->vb_dev_info
);
552 vb
->nb
.notifier_call
= virtballoon_oom_notify
;
553 vb
->nb
.priority
= VIRTBALLOON_OOM_NOTIFY_PRIORITY
;
554 err
= register_oom_notifier(&vb
->nb
);
558 #ifdef CONFIG_BALLOON_COMPACTION
559 balloon_mnt
= kern_mount(&balloon_fs
);
560 if (IS_ERR(balloon_mnt
)) {
561 err
= PTR_ERR(balloon_mnt
);
562 unregister_oom_notifier(&vb
->nb
);
566 vb
->vb_dev_info
.migratepage
= virtballoon_migratepage
;
567 vb
->vb_dev_info
.inode
= alloc_anon_inode(balloon_mnt
->mnt_sb
);
568 if (IS_ERR(vb
->vb_dev_info
.inode
)) {
569 err
= PTR_ERR(vb
->vb_dev_info
.inode
);
570 kern_unmount(balloon_mnt
);
571 unregister_oom_notifier(&vb
->nb
);
572 vb
->vb_dev_info
.inode
= NULL
;
575 vb
->vb_dev_info
.inode
->i_mapping
->a_ops
= &balloon_aops
;
578 virtio_device_ready(vdev
);
580 if (towards_target(vb
))
581 virtballoon_changed(vdev
);
585 vdev
->config
->del_vqs(vdev
);
592 static void remove_common(struct virtio_balloon
*vb
)
594 /* There might be pages left in the balloon: free them. */
595 while (vb
->num_pages
)
596 leak_balloon(vb
, vb
->num_pages
);
597 update_balloon_size(vb
);
599 /* Now we reset the device so we can clean up the queues. */
600 vb
->vdev
->config
->reset(vb
->vdev
);
602 vb
->vdev
->config
->del_vqs(vb
->vdev
);
605 static void virtballoon_remove(struct virtio_device
*vdev
)
607 struct virtio_balloon
*vb
= vdev
->priv
;
609 unregister_oom_notifier(&vb
->nb
);
611 spin_lock_irq(&vb
->stop_update_lock
);
612 vb
->stop_update
= true;
613 spin_unlock_irq(&vb
->stop_update_lock
);
614 cancel_work_sync(&vb
->update_balloon_size_work
);
615 cancel_work_sync(&vb
->update_balloon_stats_work
);
618 if (vb
->vb_dev_info
.inode
)
619 iput(vb
->vb_dev_info
.inode
);
623 #ifdef CONFIG_PM_SLEEP
624 static int virtballoon_freeze(struct virtio_device
*vdev
)
626 struct virtio_balloon
*vb
= vdev
->priv
;
629 * The workqueue is already frozen by the PM core before this
630 * function is called.
636 static int virtballoon_restore(struct virtio_device
*vdev
)
638 struct virtio_balloon
*vb
= vdev
->priv
;
641 ret
= init_vqs(vdev
->priv
);
645 virtio_device_ready(vdev
);
647 if (towards_target(vb
))
648 virtballoon_changed(vdev
);
649 update_balloon_size(vb
);
654 static unsigned int features
[] = {
655 VIRTIO_BALLOON_F_MUST_TELL_HOST
,
656 VIRTIO_BALLOON_F_STATS_VQ
,
657 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
,
660 static struct virtio_driver virtio_balloon_driver
= {
661 .feature_table
= features
,
662 .feature_table_size
= ARRAY_SIZE(features
),
663 .driver
.name
= KBUILD_MODNAME
,
664 .driver
.owner
= THIS_MODULE
,
665 .id_table
= id_table
,
666 .probe
= virtballoon_probe
,
667 .remove
= virtballoon_remove
,
668 .config_changed
= virtballoon_changed
,
669 #ifdef CONFIG_PM_SLEEP
670 .freeze
= virtballoon_freeze
,
671 .restore
= virtballoon_restore
,
675 module_virtio_driver(virtio_balloon_driver
);
676 MODULE_DEVICE_TABLE(virtio
, id_table
);
677 MODULE_DESCRIPTION("Virtio balloon driver");
678 MODULE_LICENSE("GPL");