drivers/virtio/virtio_balloon.c

   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/kthread.h>
  26 #include <linux/freezer.h>
  27 #include <linux/delay.h>
  28 #include <linux/slab.h>
  29 #include <linux/module.h>
  30 #include <linux/balloon_compaction.h>
  31
  32 /*
  33  * Balloon device works in 4K page units.  So each page is pointed to by
  34  * multiple balloon pages.  All memory counters in this driver are in balloon
  35  * page units.
  36  */
  37 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
  38 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
  39
  40 struct virtio_balloon
  41 {
  42         struct virtio_device *vdev;
  43         struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
  44
  45         /* Where the ballooning thread waits for config to change. */
  46         wait_queue_head_t config_change;
  47
  48         /* The thread servicing the balloon. */
  49         struct task_struct *thread;
  50
  51         /* Waiting for host to ack the pages we released. */
  52         wait_queue_head_t acked;
  53
  54         /* Number of balloon pages we've told the Host we're not using. */
  55         unsigned int num_pages;
  56         /*
  57          * The pages we've told the Host we're not using are enqueued
  58          * at vb_dev_info->pages list.
  59          * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
  60          * to num_pages above.
  61          */
  62         struct balloon_dev_info *vb_dev_info;
  63
  64         /* Synchronize access/update to this struct virtio_balloon elements */
  65         struct mutex balloon_lock;
  66
  67         /* The array of pfns we tell the Host about. */
  68         unsigned int num_pfns;
  69         u32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
  70
  71         /* Memory statistics */
  72         int need_stats_update;
  73         struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
  74 };
  75
  76 static struct virtio_device_id id_table[] = {
  77         { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
  78         { 0 },
  79 };
  80
  81 static u32 page_to_balloon_pfn(struct page *page)
  82 {
  83         unsigned long pfn = page_to_pfn(page);
  84
  85         BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
  86         /* Convert pfn from Linux page size to balloon page size. */
  87         return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
  88 }
  89
  90 static struct page *balloon_pfn_to_page(u32 pfn)
  91 {
  92         BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE);
  93         return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE);
  94 }
  95
  96 static void balloon_ack(struct virtqueue *vq)
  97 {
  98         struct virtio_balloon *vb = vq->vdev->priv;
  99
 100         wake_up(&vb->acked);
 101 }
 102
 103 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
 104 {
 105         struct scatterlist sg;
 106         unsigned int len;
 107
 108         sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
 109
 110         /* We should always be able to add one buffer to an empty queue. */
 111         if (virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL) < 0)
 112                 BUG();
 113         virtqueue_kick(vq);
 114
 115         /* When host has read buffer, this completes via balloon_ack */
 116         wait_event(vb->acked, virtqueue_get_buf(vq, &len));
 117 }
 118
 119 static void set_page_pfns(u32 pfns[], struct page *page)
 120 {
 121         unsigned int i;
 122
 123         /* Set balloon pfns pointing at this page.
 124          * Note that the first pfn points at start of the page. */
 125         for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
 126                 pfns[i] = page_to_balloon_pfn(page) + i;
 127 }
 128
 129 static void fill_balloon(struct virtio_balloon *vb, size_t num)
 130 {
 131         struct balloon_dev_info *vb_dev_info = vb->vb_dev_info;
 132
 133         /* We can only do one array worth at a time. */
 134         num = min(num, ARRAY_SIZE(vb->pfns));
 135
 136         mutex_lock(&vb->balloon_lock);
 137         for (vb->num_pfns = 0; vb->num_pfns < num;
 138              vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
 139                 struct page *page = balloon_page_enqueue(vb_dev_info);
 140
 141                 if (!page) {
 142                         dev_info_ratelimited(&vb->vdev->dev,
 143                                              "Out of puff! Can't get %u pages\n",
 144                                              VIRTIO_BALLOON_PAGES_PER_PAGE);
 145                         /* Sleep for at least 1/5 of a second before retry. */
 146                         msleep(200);
 147                         break;
 148                 }
 149                 set_page_pfns(vb->pfns + vb->num_pfns, page);
 150                 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
 151                 adjust_managed_page_count(page, -1);
 152         }
 153
 154         /* Did we get any? */
 155         if (vb->num_pfns != 0)
 156                 tell_host(vb, vb->inflate_vq);
 157         mutex_unlock(&vb->balloon_lock);
 158 }
 159
 160 static void release_pages_by_pfn(const u32 pfns[], unsigned int num)
 161 {
 162         unsigned int i;
 163
 164         /* Find pfns pointing at start of each page, get pages and free them. */
 165         for (i = 0; i < num; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
 166                 struct page *page = balloon_pfn_to_page(pfns[i]);
 167                 balloon_page_free(page);
 168                 adjust_managed_page_count(page, 1);
 169         }
 170 }
 171
 172 static void leak_balloon(struct virtio_balloon *vb, size_t num)
 173 {
 174         struct page *page;
 175         struct balloon_dev_info *vb_dev_info = vb->vb_dev_info;
 176
 177         /* We can only do one array worth at a time. */
 178         num = min(num, ARRAY_SIZE(vb->pfns));
 179
 180         mutex_lock(&vb->balloon_lock);
 181         for (vb->num_pfns = 0; vb->num_pfns < num;
 182              vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
 183                 page = balloon_page_dequeue(vb_dev_info);
 184                 if (!page)
 185                         break;
 186                 set_page_pfns(vb->pfns + vb->num_pfns, page);
 187                 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
 188         }
 189
 190         /*
 191          * Note that if
 192          * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
 193          * is true, we *have* to do it in this order
 194          */
 195         if (vb->num_pfns != 0)
 196                 tell_host(vb, vb->deflate_vq);
 197         mutex_unlock(&vb->balloon_lock);
 198         release_pages_by_pfn(vb->pfns, vb->num_pfns);
 199 }
 200
 201 static inline void update_stat(struct virtio_balloon *vb, int idx,
 202                                u16 tag, u64 val)
 203 {
 204         BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
 205         vb->stats[idx].tag = tag;
 206         vb->stats[idx].val = val;
 207 }
 208
 209 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
 210
 211 static void update_balloon_stats(struct virtio_balloon *vb)
 212 {
 213         unsigned long events[NR_VM_EVENT_ITEMS];
 214         struct sysinfo i;
 215         int idx = 0;
 216
 217         all_vm_events(events);
 218         si_meminfo(&i);
 219
 220         update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
 221                                 pages_to_bytes(events[PSWPIN]));
 222         update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
 223                                 pages_to_bytes(events[PSWPOUT]));
 224         update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
 225         update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
 226         update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
 227                                 pages_to_bytes(i.freeram));
 228         update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
 229                                 pages_to_bytes(i.totalram));
 230 }
 231
 232 /*
 233  * While most virtqueues communicate guest-initiated requests to the hypervisor,
 234  * the stats queue operates in reverse.  The driver initializes the virtqueue
 235  * with a single buffer.  From that point forward, all conversations consist of
 236  * a hypervisor request (a call to this function) which directs us to refill
 237  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
 238  * we notify our kthread which does the actual work via stats_handle_request().
 239  */
 240 static void stats_request(struct virtqueue *vq)
 241 {
 242         struct virtio_balloon *vb = vq->vdev->priv;
 243
 244         vb->need_stats_update = 1;
 245         wake_up(&vb->config_change);
 246 }
 247
 248 static void stats_handle_request(struct virtio_balloon *vb)
 249 {
 250         struct virtqueue *vq;
 251         struct scatterlist sg;
 252         unsigned int len;
 253
 254         vb->need_stats_update = 0;
 255         update_balloon_stats(vb);
 256
 257         vq = vb->stats_vq;
 258         if (!virtqueue_get_buf(vq, &len))
 259                 return;
 260         sg_init_one(&sg, vb->stats, sizeof(vb->stats));
 261         if (virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL) < 0)
 262                 BUG();
 263         virtqueue_kick(vq);
 264 }
 265
 266 static void virtballoon_changed(struct virtio_device *vdev)
 267 {
 268         struct virtio_balloon *vb = vdev->priv;
 269
 270         wake_up(&vb->config_change);
 271 }
 272
 273 static inline s64 towards_target(struct virtio_balloon *vb)
 274 {
 275         __le32 v;
 276         s64 target;
 277
 278         virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages, &v);
 279
 280         target = le32_to_cpu(v);
 281         return target - vb->num_pages;
 282 }
 283
 284 static void update_balloon_size(struct virtio_balloon *vb)
 285 {
 286         __le32 actual = cpu_to_le32(vb->num_pages);
 287
 288         virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
 289                       &actual);
 290 }
 291
 292 static int balloon(void *_vballoon)
 293 {
 294         struct virtio_balloon *vb = _vballoon;
 295
 296         set_freezable();
 297         while (!kthread_should_stop()) {
 298                 s64 diff;
 299
 300                 try_to_freeze();
 301                 wait_event_interruptible(vb->config_change,
 302                                          (diff = towards_target(vb)) != 0
 303                                          || vb->need_stats_update
 304                                          || kthread_should_stop()
 305                                          || freezing(current));
 306                 if (vb->need_stats_update)
 307                         stats_handle_request(vb);
 308                 if (diff > 0)
 309                         fill_balloon(vb, diff);
 310                 else if (diff < 0)
 311                         leak_balloon(vb, -diff);
 312                 update_balloon_size(vb);
 313
 314                 /*
 315                  * For large balloon changes, we could spend a lot of time
 316                  * and always have work to do.  Be nice if preempt disabled.
 317                  */
 318                 cond_resched();
 319         }
 320         return 0;
 321 }
 322
 323 static int init_vqs(struct virtio_balloon *vb)
 324 {
 325         struct virtqueue *vqs[3];
 326         vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
 327         const char *names[] = { "inflate", "deflate", "stats" };
 328         int err, nvqs;
 329
 330         /*
 331          * We expect two virtqueues: inflate and deflate, and
 332          * optionally stat.
 333          */
 334         nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
 335         err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
 336         if (err)
 337                 return err;
 338
 339         vb->inflate_vq = vqs[0];
 340         vb->deflate_vq = vqs[1];
 341         if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
 342                 struct scatterlist sg;
 343                 vb->stats_vq = vqs[2];
 344
 345                 /*
 346                  * Prime this virtqueue with one buffer so the hypervisor can
 347                  * use it to signal us later.
 348                  */
 349                 sg_init_one(&sg, vb->stats, sizeof vb->stats);
 350                 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
 351                     < 0)
 352                         BUG();
 353                 virtqueue_kick(vb->stats_vq);
 354         }
 355         return 0;
 356 }
 357
 358 static const struct address_space_operations virtio_balloon_aops;
 359 #ifdef CONFIG_BALLOON_COMPACTION
 360 /*
 361  * virtballoon_migratepage - perform the balloon page migration on behalf of
 362  *                           a compation thread.     (called under page lock)
 363  * @mapping: the page->mapping which will be assigned to the new migrated page.
 364  * @newpage: page that will replace the isolated page after migration finishes.
 365  * @page   : the isolated (old) page that is about to be migrated to newpage.
 366  * @mode   : compaction mode -- not used for balloon page migration.
 367  *
 368  * After a ballooned page gets isolated by compaction procedures, this is the
 369  * function that performs the page migration on behalf of a compaction thread
 370  * The page migration for virtio balloon is done in a simple swap fashion which
 371  * follows these two macro steps:
 372  *  1) insert newpage into vb->pages list and update the host about it;
 373  *  2) update the host about the old page removed from vb->pages list;
 374  *
 375  * This function preforms the balloon page migration task.
 376  * Called through balloon_mapping->a_ops->migratepage
 377  */
 378 static int virtballoon_migratepage(struct address_space *mapping,
 379                 struct page *newpage, struct page *page, enum migrate_mode mode)
 380 {
 381         struct balloon_dev_info *vb_dev_info = balloon_page_device(page);
 382         struct virtio_balloon *vb;
 383         unsigned long flags;
 384
 385         BUG_ON(!vb_dev_info);
 386
 387         vb = vb_dev_info->balloon_device;
 388
 389         /*
 390          * In order to avoid lock contention while migrating pages concurrently
 391          * to leak_balloon() or fill_balloon() we just give up the balloon_lock
 392          * this turn, as it is easier to retry the page migration later.
 393          * This also prevents fill_balloon() getting stuck into a mutex
 394          * recursion in the case it ends up triggering memory compaction
 395          * while it is attempting to inflate the ballon.
 396          */
 397         if (!mutex_trylock(&vb->balloon_lock))
 398                 return -EAGAIN;
 399
 400         /* balloon's page migration 1st step  -- inflate "newpage" */
 401         spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
 402         balloon_page_insert(newpage, mapping, &vb_dev_info->pages);
 403         vb_dev_info->isolated_pages--;
 404         spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
 405         vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
 406         set_page_pfns(vb->pfns, newpage);
 407         tell_host(vb, vb->inflate_vq);
 408
 409         /*
 410          * balloon's page migration 2nd step -- deflate "page"
 411          *
 412          * It's safe to delete page->lru here because this page is at
 413          * an isolated migration list, and this step is expected to happen here
 414          */
 415         balloon_page_delete(page);
 416         vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
 417         set_page_pfns(vb->pfns, page);
 418         tell_host(vb, vb->deflate_vq);
 419
 420         mutex_unlock(&vb->balloon_lock);
 421
 422         return MIGRATEPAGE_BALLOON_SUCCESS;
 423 }
 424
 425 /* define the balloon_mapping->a_ops callback to allow balloon page migration */
 426 static const struct address_space_operations virtio_balloon_aops = {
 427                         .migratepage = virtballoon_migratepage,
 428 };
 429 #endif /* CONFIG_BALLOON_COMPACTION */
 430
 431 static int virtballoon_probe(struct virtio_device *vdev)
 432 {
 433         struct virtio_balloon *vb;
 434         struct address_space *vb_mapping;
 435         struct balloon_dev_info *vb_devinfo;
 436         int err;
 437
 438         vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
 439         if (!vb) {
 440                 err = -ENOMEM;
 441                 goto out;
 442         }
 443
 444         vb->num_pages = 0;
 445         mutex_init(&vb->balloon_lock);
 446         init_waitqueue_head(&vb->config_change);
 447         init_waitqueue_head(&vb->acked);
 448         vb->vdev = vdev;
 449         vb->need_stats_update = 0;
 450
 451         vb_devinfo = balloon_devinfo_alloc(vb);
 452         if (IS_ERR(vb_devinfo)) {
 453                 err = PTR_ERR(vb_devinfo);
 454                 goto out_free_vb;
 455         }
 456
 457         vb_mapping = balloon_mapping_alloc(vb_devinfo,
 458                                            (balloon_compaction_check()) ?
 459                                            &virtio_balloon_aops : NULL);
 460         if (IS_ERR(vb_mapping)) {
 461                 /*
 462                  * IS_ERR(vb_mapping) && PTR_ERR(vb_mapping) == -EOPNOTSUPP
 463                  * This means !CONFIG_BALLOON_COMPACTION, otherwise we get off.
 464                  */
 465                 err = PTR_ERR(vb_mapping);
 466                 if (err != -EOPNOTSUPP)
 467                         goto out_free_vb_devinfo;
 468         }
 469
 470         vb->vb_dev_info = vb_devinfo;
 471
 472         err = init_vqs(vb);
 473         if (err)
 474                 goto out_free_vb_mapping;
 475
 476         vb->thread = kthread_run(balloon, vb, "vballoon");
 477         if (IS_ERR(vb->thread)) {
 478                 err = PTR_ERR(vb->thread);
 479                 goto out_del_vqs;
 480         }
 481
 482         return 0;
 483
 484 out_del_vqs:
 485         vdev->config->del_vqs(vdev);
 486 out_free_vb_mapping:
 487         balloon_mapping_free(vb_mapping);
 488 out_free_vb_devinfo:
 489         balloon_devinfo_free(vb_devinfo);
 490 out_free_vb:
 491         kfree(vb);
 492 out:
 493         return err;
 494 }
 495
 496 static void remove_common(struct virtio_balloon *vb)
 497 {
 498         /* There might be pages left in the balloon: free them. */
 499         while (vb->num_pages)
 500                 leak_balloon(vb, vb->num_pages);
 501         update_balloon_size(vb);
 502
 503         /* Now we reset the device so we can clean up the queues. */
 504         vb->vdev->config->reset(vb->vdev);
 505
 506         vb->vdev->config->del_vqs(vb->vdev);
 507 }
 508
 509 static void virtballoon_remove(struct virtio_device *vdev)
 510 {
 511         struct virtio_balloon *vb = vdev->priv;
 512
 513         kthread_stop(vb->thread);
 514         remove_common(vb);
 515         balloon_mapping_free(vb->vb_dev_info->mapping);
 516         balloon_devinfo_free(vb->vb_dev_info);
 517         kfree(vb);
 518 }
 519
 520 #ifdef CONFIG_PM_SLEEP
 521 static int virtballoon_freeze(struct virtio_device *vdev)
 522 {
 523         struct virtio_balloon *vb = vdev->priv;
 524
 525         /*
 526          * The kthread is already frozen by the PM core before this
 527          * function is called.
 528          */
 529
 530         remove_common(vb);
 531         return 0;
 532 }
 533
 534 static int virtballoon_restore(struct virtio_device *vdev)
 535 {
 536         struct virtio_balloon *vb = vdev->priv;
 537         int ret;
 538
 539         ret = init_vqs(vdev->priv);
 540         if (ret)
 541                 return ret;
 542
 543         fill_balloon(vb, towards_target(vb));
 544         update_balloon_size(vb);
 545         return 0;
 546 }
 547 #endif
 548
 549 static unsigned int features[] = {
 550         VIRTIO_BALLOON_F_MUST_TELL_HOST,
 551         VIRTIO_BALLOON_F_STATS_VQ,
 552 };
 553
 554 static struct virtio_driver virtio_balloon_driver = {
 555         .feature_table = features,
 556         .feature_table_size = ARRAY_SIZE(features),
 557         .driver.name =  KBUILD_MODNAME,
 558         .driver.owner = THIS_MODULE,
 559         .id_table =     id_table,
 560         .probe =        virtballoon_probe,
 561         .remove =       virtballoon_remove,
 562         .config_changed = virtballoon_changed,
 563 #ifdef CONFIG_PM_SLEEP
 564         .freeze =       virtballoon_freeze,
 565         .restore =      virtballoon_restore,
 566 #endif
 567 };
 568
 569 module_virtio_driver(virtio_balloon_driver);
 570 MODULE_DEVICE_TABLE(virtio, id_table);
 571 MODULE_DESCRIPTION("Virtio balloon driver");
 572 MODULE_LICENSE("GPL");