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>
34 #include <linux/magic.h>
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
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
46 static int oom_pages
= OOM_VBALLOON_DEFAULT_PAGES
;
47 module_param(oom_pages
, int, S_IRUSR
| S_IWUSR
);
48 MODULE_PARM_DESC(oom_pages
, "pages to free on OOM");
50 #ifdef CONFIG_BALLOON_COMPACTION
51 static struct vfsmount
*balloon_mnt
;
54 struct virtio_balloon
{
55 struct virtio_device
*vdev
;
56 struct virtqueue
*inflate_vq
, *deflate_vq
, *stats_vq
;
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
;
62 /* Prevent updating balloon when it is being canceled. */
63 spinlock_t stop_update_lock
;
66 /* Waiting for host to ack the pages we released. */
67 wait_queue_head_t acked
;
69 /* Number of balloon pages we've told the Host we're not using. */
70 unsigned int num_pages
;
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
77 struct balloon_dev_info vb_dev_info
;
79 /* Synchronize access/update to this struct virtio_balloon elements */
80 struct mutex balloon_lock
;
82 /* The array of pfns we tell the Host about. */
83 unsigned int num_pfns
;
84 __virtio32 pfns
[VIRTIO_BALLOON_ARRAY_PFNS_MAX
];
86 /* Memory statistics */
87 struct virtio_balloon_stat stats
[VIRTIO_BALLOON_S_NR
];
89 /* To register callback in oom notifier call chain */
90 struct notifier_block nb
;
93 static struct virtio_device_id id_table
[] = {
94 { VIRTIO_ID_BALLOON
, VIRTIO_DEV_ANY_ID
},
98 static u32
page_to_balloon_pfn(struct page
*page
)
100 unsigned long pfn
= page_to_pfn(page
);
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
;
107 static void balloon_ack(struct virtqueue
*vq
)
109 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
114 static void tell_host(struct virtio_balloon
*vb
, struct virtqueue
*vq
)
116 struct scatterlist sg
;
119 sg_init_one(&sg
, vb
->pfns
, sizeof(vb
->pfns
[0]) * vb
->num_pfns
);
121 /* We should always be able to add one buffer to an empty queue. */
122 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
125 /* When host has read buffer, this completes via balloon_ack */
126 wait_event(vb
->acked
, virtqueue_get_buf(vq
, &len
));
130 static void set_page_pfns(struct virtio_balloon
*vb
,
131 __virtio32 pfns
[], struct page
*page
)
136 * Set balloon pfns pointing at this page.
137 * Note that the first pfn points at start of the page.
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
);
144 static unsigned fill_balloon(struct virtio_balloon
*vb
, size_t num
)
146 struct balloon_dev_info
*vb_dev_info
= &vb
->vb_dev_info
;
147 unsigned num_allocated_pages
;
149 /* We can only do one array worth at a time. */
150 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
152 mutex_lock(&vb
->balloon_lock
);
153 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
154 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
155 struct page
*page
= balloon_page_enqueue(vb_dev_info
);
158 dev_info_ratelimited(&vb
->vdev
->dev
,
159 "Out of puff! Can't get %u pages\n",
160 VIRTIO_BALLOON_PAGES_PER_PAGE
);
161 /* Sleep for at least 1/5 of a second before retry. */
165 set_page_pfns(vb
, vb
->pfns
+ vb
->num_pfns
, page
);
166 vb
->num_pages
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
167 if (!virtio_has_feature(vb
->vdev
,
168 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
169 adjust_managed_page_count(page
, -1);
172 num_allocated_pages
= vb
->num_pfns
;
173 /* Did we get any? */
174 if (vb
->num_pfns
!= 0)
175 tell_host(vb
, vb
->inflate_vq
);
176 mutex_unlock(&vb
->balloon_lock
);
178 return num_allocated_pages
;
181 static void release_pages_balloon(struct virtio_balloon
*vb
,
182 struct list_head
*pages
)
184 struct page
*page
, *next
;
186 list_for_each_entry_safe(page
, next
, pages
, lru
) {
187 if (!virtio_has_feature(vb
->vdev
,
188 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
189 adjust_managed_page_count(page
, 1);
190 list_del(&page
->lru
);
191 put_page(page
); /* balloon reference */
195 static unsigned leak_balloon(struct virtio_balloon
*vb
, size_t num
)
197 unsigned num_freed_pages
;
199 struct balloon_dev_info
*vb_dev_info
= &vb
->vb_dev_info
;
202 /* We can only do one array worth at a time. */
203 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
205 mutex_lock(&vb
->balloon_lock
);
206 /* We can't release more pages than taken */
207 num
= min(num
, (size_t)vb
->num_pages
);
208 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
209 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
210 page
= balloon_page_dequeue(vb_dev_info
);
213 set_page_pfns(vb
, vb
->pfns
+ vb
->num_pfns
, page
);
214 list_add(&page
->lru
, &pages
);
215 vb
->num_pages
-= VIRTIO_BALLOON_PAGES_PER_PAGE
;
218 num_freed_pages
= vb
->num_pfns
;
221 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
222 * is true, we *have* to do it in this order
224 if (vb
->num_pfns
!= 0)
225 tell_host(vb
, vb
->deflate_vq
);
226 release_pages_balloon(vb
, &pages
);
227 mutex_unlock(&vb
->balloon_lock
);
228 return num_freed_pages
;
231 static inline void update_stat(struct virtio_balloon
*vb
, int idx
,
234 BUG_ON(idx
>= VIRTIO_BALLOON_S_NR
);
235 vb
->stats
[idx
].tag
= cpu_to_virtio16(vb
->vdev
, tag
);
236 vb
->stats
[idx
].val
= cpu_to_virtio64(vb
->vdev
, val
);
239 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
241 static unsigned int update_balloon_stats(struct virtio_balloon
*vb
)
243 unsigned long events
[NR_VM_EVENT_ITEMS
];
245 unsigned int idx
= 0;
248 all_vm_events(events
);
251 available
= si_mem_available();
253 #ifdef CONFIG_VM_EVENT_COUNTERS
254 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_IN
,
255 pages_to_bytes(events
[PSWPIN
]));
256 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_OUT
,
257 pages_to_bytes(events
[PSWPOUT
]));
258 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MAJFLT
, events
[PGMAJFAULT
]);
259 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MINFLT
, events
[PGFAULT
]);
261 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMFREE
,
262 pages_to_bytes(i
.freeram
));
263 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMTOT
,
264 pages_to_bytes(i
.totalram
));
265 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_AVAIL
,
266 pages_to_bytes(available
));
272 * While most virtqueues communicate guest-initiated requests to the hypervisor,
273 * the stats queue operates in reverse. The driver initializes the virtqueue
274 * with a single buffer. From that point forward, all conversations consist of
275 * a hypervisor request (a call to this function) which directs us to refill
276 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
277 * we delegate the job to a freezable workqueue that will do the actual work via
278 * stats_handle_request().
280 static void stats_request(struct virtqueue
*vq
)
282 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
284 spin_lock(&vb
->stop_update_lock
);
285 if (!vb
->stop_update
)
286 queue_work(system_freezable_wq
, &vb
->update_balloon_stats_work
);
287 spin_unlock(&vb
->stop_update_lock
);
290 static void stats_handle_request(struct virtio_balloon
*vb
)
292 struct virtqueue
*vq
;
293 struct scatterlist sg
;
294 unsigned int len
, num_stats
;
296 num_stats
= update_balloon_stats(vb
);
299 if (!virtqueue_get_buf(vq
, &len
))
301 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
[0]) * num_stats
);
302 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
306 static void virtballoon_changed(struct virtio_device
*vdev
)
308 struct virtio_balloon
*vb
= vdev
->priv
;
311 spin_lock_irqsave(&vb
->stop_update_lock
, flags
);
312 if (!vb
->stop_update
)
313 queue_work(system_freezable_wq
, &vb
->update_balloon_size_work
);
314 spin_unlock_irqrestore(&vb
->stop_update_lock
, flags
);
317 static inline s64
towards_target(struct virtio_balloon
*vb
)
322 virtio_cread(vb
->vdev
, struct virtio_balloon_config
, num_pages
,
325 /* Legacy balloon config space is LE, unlike all other devices. */
326 if (!virtio_has_feature(vb
->vdev
, VIRTIO_F_VERSION_1
))
327 num_pages
= le32_to_cpu((__force __le32
)num_pages
);
330 return target
- vb
->num_pages
;
333 static void update_balloon_size(struct virtio_balloon
*vb
)
335 u32 actual
= vb
->num_pages
;
337 /* Legacy balloon config space is LE, unlike all other devices. */
338 if (!virtio_has_feature(vb
->vdev
, VIRTIO_F_VERSION_1
))
339 actual
= (__force u32
)cpu_to_le32(actual
);
341 virtio_cwrite(vb
->vdev
, struct virtio_balloon_config
, actual
,
346 * virtballoon_oom_notify - release pages when system is under severe
347 * memory pressure (called from out_of_memory())
348 * @self : notifier block struct
350 * @parm : returned - number of freed pages
352 * The balancing of memory by use of the virtio balloon should not cause
353 * the termination of processes while there are pages in the balloon.
354 * If virtio balloon manages to release some memory, it will make the
355 * system return and retry the allocation that forced the OOM killer
358 static int virtballoon_oom_notify(struct notifier_block
*self
,
359 unsigned long dummy
, void *parm
)
361 struct virtio_balloon
*vb
;
362 unsigned long *freed
;
363 unsigned num_freed_pages
;
365 vb
= container_of(self
, struct virtio_balloon
, nb
);
366 if (!virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
370 num_freed_pages
= leak_balloon(vb
, oom_pages
);
371 update_balloon_size(vb
);
372 *freed
+= num_freed_pages
;
377 static void update_balloon_stats_func(struct work_struct
*work
)
379 struct virtio_balloon
*vb
;
381 vb
= container_of(work
, struct virtio_balloon
,
382 update_balloon_stats_work
);
383 stats_handle_request(vb
);
386 static void update_balloon_size_func(struct work_struct
*work
)
388 struct virtio_balloon
*vb
;
391 vb
= container_of(work
, struct virtio_balloon
,
392 update_balloon_size_work
);
393 diff
= towards_target(vb
);
396 diff
-= fill_balloon(vb
, diff
);
398 diff
+= leak_balloon(vb
, -diff
);
399 update_balloon_size(vb
);
402 queue_work(system_freezable_wq
, work
);
405 static int init_vqs(struct virtio_balloon
*vb
)
407 struct virtqueue
*vqs
[3];
408 vq_callback_t
*callbacks
[] = { balloon_ack
, balloon_ack
, stats_request
};
409 static const char * const names
[] = { "inflate", "deflate", "stats" };
413 * We expect two virtqueues: inflate and deflate, and
416 nvqs
= virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
) ? 3 : 2;
417 err
= virtio_find_vqs(vb
->vdev
, nvqs
, vqs
, callbacks
, names
, NULL
);
421 vb
->inflate_vq
= vqs
[0];
422 vb
->deflate_vq
= vqs
[1];
423 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
424 struct scatterlist sg
;
425 unsigned int num_stats
;
426 vb
->stats_vq
= vqs
[2];
429 * Prime this virtqueue with one buffer so the hypervisor can
430 * use it to signal us later (it can't be broken yet!).
432 num_stats
= update_balloon_stats(vb
);
434 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
[0]) * num_stats
);
435 if (virtqueue_add_outbuf(vb
->stats_vq
, &sg
, 1, vb
, GFP_KERNEL
)
438 virtqueue_kick(vb
->stats_vq
);
443 #ifdef CONFIG_BALLOON_COMPACTION
445 * virtballoon_migratepage - perform the balloon page migration on behalf of
446 * a compation thread. (called under page lock)
447 * @vb_dev_info: the balloon device
448 * @newpage: page that will replace the isolated page after migration finishes.
449 * @page : the isolated (old) page that is about to be migrated to newpage.
450 * @mode : compaction mode -- not used for balloon page migration.
452 * After a ballooned page gets isolated by compaction procedures, this is the
453 * function that performs the page migration on behalf of a compaction thread
454 * The page migration for virtio balloon is done in a simple swap fashion which
455 * follows these two macro steps:
456 * 1) insert newpage into vb->pages list and update the host about it;
457 * 2) update the host about the old page removed from vb->pages list;
459 * This function preforms the balloon page migration task.
460 * Called through balloon_mapping->a_ops->migratepage
462 static int virtballoon_migratepage(struct balloon_dev_info
*vb_dev_info
,
463 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
465 struct virtio_balloon
*vb
= container_of(vb_dev_info
,
466 struct virtio_balloon
, vb_dev_info
);
470 * In order to avoid lock contention while migrating pages concurrently
471 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
472 * this turn, as it is easier to retry the page migration later.
473 * This also prevents fill_balloon() getting stuck into a mutex
474 * recursion in the case it ends up triggering memory compaction
475 * while it is attempting to inflate the ballon.
477 if (!mutex_trylock(&vb
->balloon_lock
))
480 get_page(newpage
); /* balloon reference */
482 /* balloon's page migration 1st step -- inflate "newpage" */
483 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
484 balloon_page_insert(vb_dev_info
, newpage
);
485 vb_dev_info
->isolated_pages
--;
486 __count_vm_event(BALLOON_MIGRATE
);
487 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
488 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
489 set_page_pfns(vb
, vb
->pfns
, newpage
);
490 tell_host(vb
, vb
->inflate_vq
);
492 /* balloon's page migration 2nd step -- deflate "page" */
493 balloon_page_delete(page
);
494 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
495 set_page_pfns(vb
, vb
->pfns
, page
);
496 tell_host(vb
, vb
->deflate_vq
);
498 mutex_unlock(&vb
->balloon_lock
);
500 put_page(page
); /* balloon reference */
502 return MIGRATEPAGE_SUCCESS
;
505 static struct dentry
*balloon_mount(struct file_system_type
*fs_type
,
506 int flags
, const char *dev_name
, void *data
)
508 static const struct dentry_operations ops
= {
509 .d_dname
= simple_dname
,
512 return mount_pseudo(fs_type
, "balloon-kvm:", NULL
, &ops
,
516 static struct file_system_type balloon_fs
= {
517 .name
= "balloon-kvm",
518 .mount
= balloon_mount
,
519 .kill_sb
= kill_anon_super
,
522 #endif /* CONFIG_BALLOON_COMPACTION */
524 static int virtballoon_probe(struct virtio_device
*vdev
)
526 struct virtio_balloon
*vb
;
529 if (!vdev
->config
->get
) {
530 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
535 vdev
->priv
= vb
= kmalloc(sizeof(*vb
), GFP_KERNEL
);
541 INIT_WORK(&vb
->update_balloon_stats_work
, update_balloon_stats_func
);
542 INIT_WORK(&vb
->update_balloon_size_work
, update_balloon_size_func
);
543 spin_lock_init(&vb
->stop_update_lock
);
544 vb
->stop_update
= false;
546 mutex_init(&vb
->balloon_lock
);
547 init_waitqueue_head(&vb
->acked
);
550 balloon_devinfo_init(&vb
->vb_dev_info
);
556 vb
->nb
.notifier_call
= virtballoon_oom_notify
;
557 vb
->nb
.priority
= VIRTBALLOON_OOM_NOTIFY_PRIORITY
;
558 err
= register_oom_notifier(&vb
->nb
);
562 #ifdef CONFIG_BALLOON_COMPACTION
563 balloon_mnt
= kern_mount(&balloon_fs
);
564 if (IS_ERR(balloon_mnt
)) {
565 err
= PTR_ERR(balloon_mnt
);
566 unregister_oom_notifier(&vb
->nb
);
570 vb
->vb_dev_info
.migratepage
= virtballoon_migratepage
;
571 vb
->vb_dev_info
.inode
= alloc_anon_inode(balloon_mnt
->mnt_sb
);
572 if (IS_ERR(vb
->vb_dev_info
.inode
)) {
573 err
= PTR_ERR(vb
->vb_dev_info
.inode
);
574 kern_unmount(balloon_mnt
);
575 unregister_oom_notifier(&vb
->nb
);
576 vb
->vb_dev_info
.inode
= NULL
;
579 vb
->vb_dev_info
.inode
->i_mapping
->a_ops
= &balloon_aops
;
582 virtio_device_ready(vdev
);
584 if (towards_target(vb
))
585 virtballoon_changed(vdev
);
589 vdev
->config
->del_vqs(vdev
);
596 static void remove_common(struct virtio_balloon
*vb
)
598 /* There might be pages left in the balloon: free them. */
599 while (vb
->num_pages
)
600 leak_balloon(vb
, vb
->num_pages
);
601 update_balloon_size(vb
);
603 /* Now we reset the device so we can clean up the queues. */
604 vb
->vdev
->config
->reset(vb
->vdev
);
606 vb
->vdev
->config
->del_vqs(vb
->vdev
);
609 static void virtballoon_remove(struct virtio_device
*vdev
)
611 struct virtio_balloon
*vb
= vdev
->priv
;
613 unregister_oom_notifier(&vb
->nb
);
615 spin_lock_irq(&vb
->stop_update_lock
);
616 vb
->stop_update
= true;
617 spin_unlock_irq(&vb
->stop_update_lock
);
618 cancel_work_sync(&vb
->update_balloon_size_work
);
619 cancel_work_sync(&vb
->update_balloon_stats_work
);
622 #ifdef CONFIG_BALLOON_COMPACTION
623 if (vb
->vb_dev_info
.inode
)
624 iput(vb
->vb_dev_info
.inode
);
626 kern_unmount(balloon_mnt
);
631 #ifdef CONFIG_PM_SLEEP
632 static int virtballoon_freeze(struct virtio_device
*vdev
)
634 struct virtio_balloon
*vb
= vdev
->priv
;
637 * The workqueue is already frozen by the PM core before this
638 * function is called.
644 static int virtballoon_restore(struct virtio_device
*vdev
)
646 struct virtio_balloon
*vb
= vdev
->priv
;
649 ret
= init_vqs(vdev
->priv
);
653 virtio_device_ready(vdev
);
655 if (towards_target(vb
))
656 virtballoon_changed(vdev
);
657 update_balloon_size(vb
);
662 static int virtballoon_validate(struct virtio_device
*vdev
)
664 __virtio_clear_bit(vdev
, VIRTIO_F_IOMMU_PLATFORM
);
668 static unsigned int features
[] = {
669 VIRTIO_BALLOON_F_MUST_TELL_HOST
,
670 VIRTIO_BALLOON_F_STATS_VQ
,
671 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
,
674 static struct virtio_driver virtio_balloon_driver
= {
675 .feature_table
= features
,
676 .feature_table_size
= ARRAY_SIZE(features
),
677 .driver
.name
= KBUILD_MODNAME
,
678 .driver
.owner
= THIS_MODULE
,
679 .id_table
= id_table
,
680 .validate
= virtballoon_validate
,
681 .probe
= virtballoon_probe
,
682 .remove
= virtballoon_remove
,
683 .config_changed
= virtballoon_changed
,
684 #ifdef CONFIG_PM_SLEEP
685 .freeze
= virtballoon_freeze
,
686 .restore
= virtballoon_restore
,
690 module_virtio_driver(virtio_balloon_driver
);
691 MODULE_DEVICE_TABLE(virtio
, id_table
);
692 MODULE_DESCRIPTION("Virtio balloon driver");
693 MODULE_LICENSE("GPL");