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>
35 * Balloon device works in 4K page units. So each page is pointed to by
36 * multiple balloon pages. All memory counters in this driver are in balloon
39 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
40 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
41 #define OOM_VBALLOON_DEFAULT_PAGES 256
42 #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
44 static int oom_pages
= OOM_VBALLOON_DEFAULT_PAGES
;
45 module_param(oom_pages
, int, S_IRUSR
| S_IWUSR
);
46 MODULE_PARM_DESC(oom_pages
, "pages to free on OOM");
48 struct virtio_balloon
{
49 struct virtio_device
*vdev
;
50 struct virtqueue
*inflate_vq
, *deflate_vq
, *stats_vq
;
52 /* The balloon servicing is delegated to a freezable workqueue. */
53 struct work_struct update_balloon_stats_work
;
54 struct work_struct update_balloon_size_work
;
56 /* Prevent updating balloon when it is being canceled. */
57 spinlock_t stop_update_lock
;
60 /* Waiting for host to ack the pages we released. */
61 wait_queue_head_t acked
;
63 /* Number of balloon pages we've told the Host we're not using. */
64 unsigned int num_pages
;
66 * The pages we've told the Host we're not using are enqueued
67 * at vb_dev_info->pages list.
68 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
71 struct balloon_dev_info vb_dev_info
;
73 /* Synchronize access/update to this struct virtio_balloon elements */
74 struct mutex balloon_lock
;
76 /* The array of pfns we tell the Host about. */
77 unsigned int num_pfns
;
78 u32 pfns
[VIRTIO_BALLOON_ARRAY_PFNS_MAX
];
80 /* Memory statistics */
81 struct virtio_balloon_stat stats
[VIRTIO_BALLOON_S_NR
];
83 /* To register callback in oom notifier call chain */
84 struct notifier_block nb
;
87 static struct virtio_device_id id_table
[] = {
88 { VIRTIO_ID_BALLOON
, VIRTIO_DEV_ANY_ID
},
92 static u32
page_to_balloon_pfn(struct page
*page
)
94 unsigned long pfn
= page_to_pfn(page
);
96 BUILD_BUG_ON(PAGE_SHIFT
< VIRTIO_BALLOON_PFN_SHIFT
);
97 /* Convert pfn from Linux page size to balloon page size. */
98 return pfn
* VIRTIO_BALLOON_PAGES_PER_PAGE
;
101 static struct page
*balloon_pfn_to_page(u32 pfn
)
103 BUG_ON(pfn
% VIRTIO_BALLOON_PAGES_PER_PAGE
);
104 return pfn_to_page(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(u32 pfns
[], struct page
*page
)
134 /* Set balloon pfns pointing at this page.
135 * Note that the first pfn points at start of the page. */
136 for (i
= 0; i
< VIRTIO_BALLOON_PAGES_PER_PAGE
; i
++)
137 pfns
[i
] = page_to_balloon_pfn(page
) + i
;
140 static unsigned fill_balloon(struct virtio_balloon
*vb
, size_t num
)
142 struct balloon_dev_info
*vb_dev_info
= &vb
->vb_dev_info
;
143 unsigned num_allocated_pages
;
145 /* We can only do one array worth at a time. */
146 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
148 mutex_lock(&vb
->balloon_lock
);
149 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
150 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
151 struct page
*page
= balloon_page_enqueue(vb_dev_info
);
154 dev_info_ratelimited(&vb
->vdev
->dev
,
155 "Out of puff! Can't get %u pages\n",
156 VIRTIO_BALLOON_PAGES_PER_PAGE
);
157 /* Sleep for at least 1/5 of a second before retry. */
161 set_page_pfns(vb
->pfns
+ vb
->num_pfns
, page
);
162 vb
->num_pages
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
163 if (!virtio_has_feature(vb
->vdev
,
164 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
165 adjust_managed_page_count(page
, -1);
168 num_allocated_pages
= vb
->num_pfns
;
169 /* Did we get any? */
170 if (vb
->num_pfns
!= 0)
171 tell_host(vb
, vb
->inflate_vq
);
172 mutex_unlock(&vb
->balloon_lock
);
174 return num_allocated_pages
;
177 static void release_pages_balloon(struct virtio_balloon
*vb
)
181 /* Find pfns pointing at start of each page, get pages and free them. */
182 for (i
= 0; i
< vb
->num_pfns
; i
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
183 struct page
*page
= balloon_pfn_to_page(vb
->pfns
[i
]);
184 if (!virtio_has_feature(vb
->vdev
,
185 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
186 adjust_managed_page_count(page
, 1);
187 put_page(page
); /* balloon reference */
191 static unsigned leak_balloon(struct virtio_balloon
*vb
, size_t num
)
193 unsigned num_freed_pages
;
195 struct balloon_dev_info
*vb_dev_info
= &vb
->vb_dev_info
;
197 /* We can only do one array worth at a time. */
198 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
200 mutex_lock(&vb
->balloon_lock
);
201 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
202 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
203 page
= balloon_page_dequeue(vb_dev_info
);
206 set_page_pfns(vb
->pfns
+ vb
->num_pfns
, page
);
207 vb
->num_pages
-= VIRTIO_BALLOON_PAGES_PER_PAGE
;
210 num_freed_pages
= vb
->num_pfns
;
213 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
214 * is true, we *have* to do it in this order
216 if (vb
->num_pfns
!= 0)
217 tell_host(vb
, vb
->deflate_vq
);
218 release_pages_balloon(vb
);
219 mutex_unlock(&vb
->balloon_lock
);
220 return num_freed_pages
;
223 static inline void update_stat(struct virtio_balloon
*vb
, int idx
,
226 BUG_ON(idx
>= VIRTIO_BALLOON_S_NR
);
227 vb
->stats
[idx
].tag
= cpu_to_virtio16(vb
->vdev
, tag
);
228 vb
->stats
[idx
].val
= cpu_to_virtio64(vb
->vdev
, val
);
231 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
233 static void update_balloon_stats(struct virtio_balloon
*vb
)
235 unsigned long events
[NR_VM_EVENT_ITEMS
];
240 all_vm_events(events
);
243 available
= si_mem_available();
245 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_IN
,
246 pages_to_bytes(events
[PSWPIN
]));
247 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_OUT
,
248 pages_to_bytes(events
[PSWPOUT
]));
249 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MAJFLT
, events
[PGMAJFAULT
]);
250 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MINFLT
, events
[PGFAULT
]);
251 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMFREE
,
252 pages_to_bytes(i
.freeram
));
253 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMTOT
,
254 pages_to_bytes(i
.totalram
));
255 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_AVAIL
,
256 pages_to_bytes(available
));
260 * While most virtqueues communicate guest-initiated requests to the hypervisor,
261 * the stats queue operates in reverse. The driver initializes the virtqueue
262 * with a single buffer. From that point forward, all conversations consist of
263 * a hypervisor request (a call to this function) which directs us to refill
264 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
265 * we delegate the job to a freezable workqueue that will do the actual work via
266 * stats_handle_request().
268 static void stats_request(struct virtqueue
*vq
)
270 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
272 spin_lock(&vb
->stop_update_lock
);
273 if (!vb
->stop_update
)
274 queue_work(system_freezable_wq
, &vb
->update_balloon_stats_work
);
275 spin_unlock(&vb
->stop_update_lock
);
278 static void stats_handle_request(struct virtio_balloon
*vb
)
280 struct virtqueue
*vq
;
281 struct scatterlist sg
;
284 update_balloon_stats(vb
);
287 if (!virtqueue_get_buf(vq
, &len
))
289 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
));
290 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
294 static void virtballoon_changed(struct virtio_device
*vdev
)
296 struct virtio_balloon
*vb
= vdev
->priv
;
299 spin_lock_irqsave(&vb
->stop_update_lock
, flags
);
300 if (!vb
->stop_update
)
301 queue_work(system_freezable_wq
, &vb
->update_balloon_size_work
);
302 spin_unlock_irqrestore(&vb
->stop_update_lock
, flags
);
305 static inline s64
towards_target(struct virtio_balloon
*vb
)
310 virtio_cread(vb
->vdev
, struct virtio_balloon_config
, num_pages
,
313 /* Legacy balloon config space is LE, unlike all other devices. */
314 if (!virtio_has_feature(vb
->vdev
, VIRTIO_F_VERSION_1
))
315 num_pages
= le32_to_cpu((__force __le32
)num_pages
);
318 return target
- vb
->num_pages
;
321 static void update_balloon_size(struct virtio_balloon
*vb
)
323 u32 actual
= vb
->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 actual
= (__force u32
)cpu_to_le32(actual
);
329 virtio_cwrite(vb
->vdev
, struct virtio_balloon_config
, actual
,
334 * virtballoon_oom_notify - release pages when system is under severe
335 * memory pressure (called from out_of_memory())
336 * @self : notifier block struct
338 * @parm : returned - number of freed pages
340 * The balancing of memory by use of the virtio balloon should not cause
341 * the termination of processes while there are pages in the balloon.
342 * If virtio balloon manages to release some memory, it will make the
343 * system return and retry the allocation that forced the OOM killer
346 static int virtballoon_oom_notify(struct notifier_block
*self
,
347 unsigned long dummy
, void *parm
)
349 struct virtio_balloon
*vb
;
350 unsigned long *freed
;
351 unsigned num_freed_pages
;
353 vb
= container_of(self
, struct virtio_balloon
, nb
);
354 if (!virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
358 num_freed_pages
= leak_balloon(vb
, oom_pages
);
359 update_balloon_size(vb
);
360 *freed
+= num_freed_pages
;
365 static void update_balloon_stats_func(struct work_struct
*work
)
367 struct virtio_balloon
*vb
;
369 vb
= container_of(work
, struct virtio_balloon
,
370 update_balloon_stats_work
);
371 stats_handle_request(vb
);
374 static void update_balloon_size_func(struct work_struct
*work
)
376 struct virtio_balloon
*vb
;
379 vb
= container_of(work
, struct virtio_balloon
,
380 update_balloon_size_work
);
381 diff
= towards_target(vb
);
384 diff
-= fill_balloon(vb
, diff
);
386 diff
+= leak_balloon(vb
, -diff
);
387 update_balloon_size(vb
);
390 queue_work(system_freezable_wq
, work
);
393 static int init_vqs(struct virtio_balloon
*vb
)
395 struct virtqueue
*vqs
[3];
396 vq_callback_t
*callbacks
[] = { balloon_ack
, balloon_ack
, stats_request
};
397 static const char * const names
[] = { "inflate", "deflate", "stats" };
401 * We expect two virtqueues: inflate and deflate, and
404 nvqs
= virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
) ? 3 : 2;
405 err
= vb
->vdev
->config
->find_vqs(vb
->vdev
, nvqs
, vqs
, callbacks
, names
);
409 vb
->inflate_vq
= vqs
[0];
410 vb
->deflate_vq
= vqs
[1];
411 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
412 struct scatterlist sg
;
413 vb
->stats_vq
= vqs
[2];
416 * Prime this virtqueue with one buffer so the hypervisor can
417 * use it to signal us later (it can't be broken yet!).
419 sg_init_one(&sg
, vb
->stats
, sizeof vb
->stats
);
420 if (virtqueue_add_outbuf(vb
->stats_vq
, &sg
, 1, vb
, GFP_KERNEL
)
423 virtqueue_kick(vb
->stats_vq
);
428 #ifdef CONFIG_BALLOON_COMPACTION
430 * virtballoon_migratepage - perform the balloon page migration on behalf of
431 * a compation thread. (called under page lock)
432 * @vb_dev_info: the balloon device
433 * @newpage: page that will replace the isolated page after migration finishes.
434 * @page : the isolated (old) page that is about to be migrated to newpage.
435 * @mode : compaction mode -- not used for balloon page migration.
437 * After a ballooned page gets isolated by compaction procedures, this is the
438 * function that performs the page migration on behalf of a compaction thread
439 * The page migration for virtio balloon is done in a simple swap fashion which
440 * follows these two macro steps:
441 * 1) insert newpage into vb->pages list and update the host about it;
442 * 2) update the host about the old page removed from vb->pages list;
444 * This function preforms the balloon page migration task.
445 * Called through balloon_mapping->a_ops->migratepage
447 static int virtballoon_migratepage(struct balloon_dev_info
*vb_dev_info
,
448 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
450 struct virtio_balloon
*vb
= container_of(vb_dev_info
,
451 struct virtio_balloon
, vb_dev_info
);
455 * In order to avoid lock contention while migrating pages concurrently
456 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
457 * this turn, as it is easier to retry the page migration later.
458 * This also prevents fill_balloon() getting stuck into a mutex
459 * recursion in the case it ends up triggering memory compaction
460 * while it is attempting to inflate the ballon.
462 if (!mutex_trylock(&vb
->balloon_lock
))
465 get_page(newpage
); /* balloon reference */
467 /* balloon's page migration 1st step -- inflate "newpage" */
468 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
469 balloon_page_insert(vb_dev_info
, newpage
);
470 vb_dev_info
->isolated_pages
--;
471 __count_vm_event(BALLOON_MIGRATE
);
472 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
473 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
474 set_page_pfns(vb
->pfns
, newpage
);
475 tell_host(vb
, vb
->inflate_vq
);
477 /* balloon's page migration 2nd step -- deflate "page" */
478 balloon_page_delete(page
);
479 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
480 set_page_pfns(vb
->pfns
, page
);
481 tell_host(vb
, vb
->deflate_vq
);
483 mutex_unlock(&vb
->balloon_lock
);
485 put_page(page
); /* balloon reference */
487 return MIGRATEPAGE_SUCCESS
;
489 #endif /* CONFIG_BALLOON_COMPACTION */
491 static int virtballoon_probe(struct virtio_device
*vdev
)
493 struct virtio_balloon
*vb
;
496 if (!vdev
->config
->get
) {
497 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
502 vdev
->priv
= vb
= kmalloc(sizeof(*vb
), GFP_KERNEL
);
508 INIT_WORK(&vb
->update_balloon_stats_work
, update_balloon_stats_func
);
509 INIT_WORK(&vb
->update_balloon_size_work
, update_balloon_size_func
);
510 spin_lock_init(&vb
->stop_update_lock
);
511 vb
->stop_update
= false;
513 mutex_init(&vb
->balloon_lock
);
514 init_waitqueue_head(&vb
->acked
);
517 balloon_devinfo_init(&vb
->vb_dev_info
);
518 #ifdef CONFIG_BALLOON_COMPACTION
519 vb
->vb_dev_info
.migratepage
= virtballoon_migratepage
;
526 vb
->nb
.notifier_call
= virtballoon_oom_notify
;
527 vb
->nb
.priority
= VIRTBALLOON_OOM_NOTIFY_PRIORITY
;
528 err
= register_oom_notifier(&vb
->nb
);
532 virtio_device_ready(vdev
);
537 vdev
->config
->del_vqs(vdev
);
544 static void remove_common(struct virtio_balloon
*vb
)
546 /* There might be pages left in the balloon: free them. */
547 while (vb
->num_pages
)
548 leak_balloon(vb
, vb
->num_pages
);
549 update_balloon_size(vb
);
551 /* Now we reset the device so we can clean up the queues. */
552 vb
->vdev
->config
->reset(vb
->vdev
);
554 vb
->vdev
->config
->del_vqs(vb
->vdev
);
557 static void virtballoon_remove(struct virtio_device
*vdev
)
559 struct virtio_balloon
*vb
= vdev
->priv
;
561 unregister_oom_notifier(&vb
->nb
);
563 spin_lock_irq(&vb
->stop_update_lock
);
564 vb
->stop_update
= true;
565 spin_unlock_irq(&vb
->stop_update_lock
);
566 cancel_work_sync(&vb
->update_balloon_size_work
);
567 cancel_work_sync(&vb
->update_balloon_stats_work
);
573 #ifdef CONFIG_PM_SLEEP
574 static int virtballoon_freeze(struct virtio_device
*vdev
)
576 struct virtio_balloon
*vb
= vdev
->priv
;
579 * The workqueue is already frozen by the PM core before this
580 * function is called.
586 static int virtballoon_restore(struct virtio_device
*vdev
)
588 struct virtio_balloon
*vb
= vdev
->priv
;
591 ret
= init_vqs(vdev
->priv
);
595 virtio_device_ready(vdev
);
597 if (towards_target(vb
))
598 virtballoon_changed(vdev
);
599 update_balloon_size(vb
);
604 static unsigned int features
[] = {
605 VIRTIO_BALLOON_F_MUST_TELL_HOST
,
606 VIRTIO_BALLOON_F_STATS_VQ
,
607 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
,
610 static struct virtio_driver virtio_balloon_driver
= {
611 .feature_table
= features
,
612 .feature_table_size
= ARRAY_SIZE(features
),
613 .driver
.name
= KBUILD_MODNAME
,
614 .driver
.owner
= THIS_MODULE
,
615 .id_table
= id_table
,
616 .probe
= virtballoon_probe
,
617 .remove
= virtballoon_remove
,
618 .config_changed
= virtballoon_changed
,
619 #ifdef CONFIG_PM_SLEEP
620 .freeze
= virtballoon_freeze
,
621 .restore
= virtballoon_restore
,
625 module_virtio_driver(virtio_balloon_driver
);
626 MODULE_DEVICE_TABLE(virtio
, id_table
);
627 MODULE_DESCRIPTION("Virtio balloon driver");
628 MODULE_LICENSE("GPL");