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 34 /* 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 37 * page units. 38 */ 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 43 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"); 47 48 struct virtio_balloon { 49 struct virtio_device *vdev; 50 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq; 51 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; 55 56 /* Prevent updating balloon when it is being canceled. */ 57 spinlock_t stop_update_lock; 58 bool stop_update; 59 60 /* Waiting for host to ack the pages we released. */ 61 wait_queue_head_t acked; 62 63 /* Number of balloon pages we've told the Host we're not using. */ 64 unsigned int num_pages; 65 /* 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 69 * to num_pages above. 70 */ 71 struct balloon_dev_info vb_dev_info; 72 73 /* Synchronize access/update to this struct virtio_balloon elements */ 74 struct mutex balloon_lock; 75 76 /* The array of pfns we tell the Host about. */ 77 unsigned int num_pfns; 78 __virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX]; 79 80 /* Memory statistics */ 81 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR]; 82 83 /* To register callback in oom notifier call chain */ 84 struct notifier_block nb; 85 }; 86 87 static struct virtio_device_id id_table[] = { 88 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID }, 89 { 0 }, 90 }; 91 92 static u32 page_to_balloon_pfn(struct page *page) 93 { 94 unsigned long pfn = page_to_pfn(page); 95 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; 99 } 100 101 static struct page *balloon_pfn_to_page(u32 pfn) 102 { 103 BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE); 104 return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE); 105 } 106 107 static void balloon_ack(struct virtqueue *vq) 108 { 109 struct virtio_balloon *vb = vq->vdev->priv; 110 111 wake_up(&vb->acked); 112 } 113 114 static 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 } 129 130 static void set_page_pfns(struct virtio_balloon *vb, 131 __virtio32 pfns[], struct page *page) 132 { 133 unsigned int i; 134 135 /* Set balloon pfns pointing at this page. 136 * Note that the first pfn points at start of the page. */ 137 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++) 138 pfns[i] = cpu_to_virtio32(vb->vdev, 139 page_to_balloon_pfn(page) + i); 140 } 141 142 static unsigned fill_balloon(struct virtio_balloon *vb, size_t num) 143 { 144 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info; 145 unsigned num_allocated_pages; 146 147 /* We can only do one array worth at a time. */ 148 num = min(num, ARRAY_SIZE(vb->pfns)); 149 150 mutex_lock(&vb->balloon_lock); 151 for (vb->num_pfns = 0; vb->num_pfns < num; 152 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { 153 struct page *page = balloon_page_enqueue(vb_dev_info); 154 155 if (!page) { 156 dev_info_ratelimited(&vb->vdev->dev, 157 "Out of puff! Can't get %u pages\n", 158 VIRTIO_BALLOON_PAGES_PER_PAGE); 159 /* Sleep for at least 1/5 of a second before retry. */ 160 msleep(200); 161 break; 162 } 163 set_page_pfns(vb, vb->pfns + vb->num_pfns, page); 164 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE; 165 if (!virtio_has_feature(vb->vdev, 166 VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 167 adjust_managed_page_count(page, -1); 168 } 169 170 num_allocated_pages = vb->num_pfns; 171 /* Did we get any? */ 172 if (vb->num_pfns != 0) 173 tell_host(vb, vb->inflate_vq); 174 mutex_unlock(&vb->balloon_lock); 175 176 return num_allocated_pages; 177 } 178 179 static void release_pages_balloon(struct virtio_balloon *vb) 180 { 181 unsigned int i; 182 struct page *page; 183 184 /* Find pfns pointing at start of each page, get pages and free them. */ 185 for (i = 0; i < vb->num_pfns; i += VIRTIO_BALLOON_PAGES_PER_PAGE) { 186 page = balloon_pfn_to_page(virtio32_to_cpu(vb->vdev, 187 vb->pfns[i])); 188 if (!virtio_has_feature(vb->vdev, 189 VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 190 adjust_managed_page_count(page, 1); 191 put_page(page); /* balloon reference */ 192 } 193 } 194 195 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num) 196 { 197 unsigned num_freed_pages; 198 struct page *page; 199 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info; 200 201 /* We can only do one array worth at a time. */ 202 num = min(num, ARRAY_SIZE(vb->pfns)); 203 204 mutex_lock(&vb->balloon_lock); 205 for (vb->num_pfns = 0; vb->num_pfns < num; 206 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { 207 page = balloon_page_dequeue(vb_dev_info); 208 if (!page) 209 break; 210 set_page_pfns(vb, vb->pfns + vb->num_pfns, page); 211 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE; 212 } 213 214 num_freed_pages = vb->num_pfns; 215 /* 216 * Note that if 217 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST); 218 * is true, we *have* to do it in this order 219 */ 220 if (vb->num_pfns != 0) 221 tell_host(vb, vb->deflate_vq); 222 release_pages_balloon(vb); 223 mutex_unlock(&vb->balloon_lock); 224 return num_freed_pages; 225 } 226 227 static inline void update_stat(struct virtio_balloon *vb, int idx, 228 u16 tag, u64 val) 229 { 230 BUG_ON(idx >= VIRTIO_BALLOON_S_NR); 231 vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag); 232 vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val); 233 } 234 235 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT) 236 237 static void update_balloon_stats(struct virtio_balloon *vb) 238 { 239 unsigned long events[NR_VM_EVENT_ITEMS]; 240 struct sysinfo i; 241 int idx = 0; 242 long available; 243 244 all_vm_events(events); 245 si_meminfo(&i); 246 247 available = si_mem_available(); 248 249 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN, 250 pages_to_bytes(events[PSWPIN])); 251 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT, 252 pages_to_bytes(events[PSWPOUT])); 253 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]); 254 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]); 255 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE, 256 pages_to_bytes(i.freeram)); 257 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT, 258 pages_to_bytes(i.totalram)); 259 update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL, 260 pages_to_bytes(available)); 261 } 262 263 /* 264 * While most virtqueues communicate guest-initiated requests to the hypervisor, 265 * the stats queue operates in reverse. The driver initializes the virtqueue 266 * with a single buffer. From that point forward, all conversations consist of 267 * a hypervisor request (a call to this function) which directs us to refill 268 * the virtqueue with a fresh stats buffer. Since stats collection can sleep, 269 * we delegate the job to a freezable workqueue that will do the actual work via 270 * stats_handle_request(). 271 */ 272 static void stats_request(struct virtqueue *vq) 273 { 274 struct virtio_balloon *vb = vq->vdev->priv; 275 276 spin_lock(&vb->stop_update_lock); 277 if (!vb->stop_update) 278 queue_work(system_freezable_wq, &vb->update_balloon_stats_work); 279 spin_unlock(&vb->stop_update_lock); 280 } 281 282 static void stats_handle_request(struct virtio_balloon *vb) 283 { 284 struct virtqueue *vq; 285 struct scatterlist sg; 286 unsigned int len; 287 288 update_balloon_stats(vb); 289 290 vq = vb->stats_vq; 291 if (!virtqueue_get_buf(vq, &len)) 292 return; 293 sg_init_one(&sg, vb->stats, sizeof(vb->stats)); 294 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); 295 virtqueue_kick(vq); 296 } 297 298 static void virtballoon_changed(struct virtio_device *vdev) 299 { 300 struct virtio_balloon *vb = vdev->priv; 301 unsigned long flags; 302 303 spin_lock_irqsave(&vb->stop_update_lock, flags); 304 if (!vb->stop_update) 305 queue_work(system_freezable_wq, &vb->update_balloon_size_work); 306 spin_unlock_irqrestore(&vb->stop_update_lock, flags); 307 } 308 309 static inline s64 towards_target(struct virtio_balloon *vb) 310 { 311 s64 target; 312 u32 num_pages; 313 314 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages, 315 &num_pages); 316 317 /* Legacy balloon config space is LE, unlike all other devices. */ 318 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1)) 319 num_pages = le32_to_cpu((__force __le32)num_pages); 320 321 target = num_pages; 322 return target - vb->num_pages; 323 } 324 325 static void update_balloon_size(struct virtio_balloon *vb) 326 { 327 u32 actual = vb->num_pages; 328 329 /* Legacy balloon config space is LE, unlike all other devices. */ 330 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1)) 331 actual = (__force u32)cpu_to_le32(actual); 332 333 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual, 334 &actual); 335 } 336 337 /* 338 * virtballoon_oom_notify - release pages when system is under severe 339 * memory pressure (called from out_of_memory()) 340 * @self : notifier block struct 341 * @dummy: not used 342 * @parm : returned - number of freed pages 343 * 344 * The balancing of memory by use of the virtio balloon should not cause 345 * the termination of processes while there are pages in the balloon. 346 * If virtio balloon manages to release some memory, it will make the 347 * system return and retry the allocation that forced the OOM killer 348 * to run. 349 */ 350 static int virtballoon_oom_notify(struct notifier_block *self, 351 unsigned long dummy, void *parm) 352 { 353 struct virtio_balloon *vb; 354 unsigned long *freed; 355 unsigned num_freed_pages; 356 357 vb = container_of(self, struct virtio_balloon, nb); 358 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 359 return NOTIFY_OK; 360 361 freed = parm; 362 num_freed_pages = leak_balloon(vb, oom_pages); 363 update_balloon_size(vb); 364 *freed += num_freed_pages; 365 366 return NOTIFY_OK; 367 } 368 369 static void update_balloon_stats_func(struct work_struct *work) 370 { 371 struct virtio_balloon *vb; 372 373 vb = container_of(work, struct virtio_balloon, 374 update_balloon_stats_work); 375 stats_handle_request(vb); 376 } 377 378 static void update_balloon_size_func(struct work_struct *work) 379 { 380 struct virtio_balloon *vb; 381 s64 diff; 382 383 vb = container_of(work, struct virtio_balloon, 384 update_balloon_size_work); 385 diff = towards_target(vb); 386 387 if (diff > 0) 388 diff -= fill_balloon(vb, diff); 389 else if (diff < 0) 390 diff += leak_balloon(vb, -diff); 391 update_balloon_size(vb); 392 393 if (diff) 394 queue_work(system_freezable_wq, work); 395 } 396 397 static int init_vqs(struct virtio_balloon *vb) 398 { 399 struct virtqueue *vqs[3]; 400 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request }; 401 static const char * const names[] = { "inflate", "deflate", "stats" }; 402 int err, nvqs; 403 404 /* 405 * We expect two virtqueues: inflate and deflate, and 406 * optionally stat. 407 */ 408 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2; 409 err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names); 410 if (err) 411 return err; 412 413 vb->inflate_vq = vqs[0]; 414 vb->deflate_vq = vqs[1]; 415 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { 416 struct scatterlist sg; 417 vb->stats_vq = vqs[2]; 418 419 /* 420 * Prime this virtqueue with one buffer so the hypervisor can 421 * use it to signal us later (it can't be broken yet!). 422 */ 423 sg_init_one(&sg, vb->stats, sizeof vb->stats); 424 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL) 425 < 0) 426 BUG(); 427 virtqueue_kick(vb->stats_vq); 428 } 429 return 0; 430 } 431 432 #ifdef CONFIG_BALLOON_COMPACTION 433 /* 434 * virtballoon_migratepage - perform the balloon page migration on behalf of 435 * a compation thread. (called under page lock) 436 * @vb_dev_info: the balloon device 437 * @newpage: page that will replace the isolated page after migration finishes. 438 * @page : the isolated (old) page that is about to be migrated to newpage. 439 * @mode : compaction mode -- not used for balloon page migration. 440 * 441 * After a ballooned page gets isolated by compaction procedures, this is the 442 * function that performs the page migration on behalf of a compaction thread 443 * The page migration for virtio balloon is done in a simple swap fashion which 444 * follows these two macro steps: 445 * 1) insert newpage into vb->pages list and update the host about it; 446 * 2) update the host about the old page removed from vb->pages list; 447 * 448 * This function preforms the balloon page migration task. 449 * Called through balloon_mapping->a_ops->migratepage 450 */ 451 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info, 452 struct page *newpage, struct page *page, enum migrate_mode mode) 453 { 454 struct virtio_balloon *vb = container_of(vb_dev_info, 455 struct virtio_balloon, vb_dev_info); 456 unsigned long flags; 457 458 /* 459 * In order to avoid lock contention while migrating pages concurrently 460 * to leak_balloon() or fill_balloon() we just give up the balloon_lock 461 * this turn, as it is easier to retry the page migration later. 462 * This also prevents fill_balloon() getting stuck into a mutex 463 * recursion in the case it ends up triggering memory compaction 464 * while it is attempting to inflate the ballon. 465 */ 466 if (!mutex_trylock(&vb->balloon_lock)) 467 return -EAGAIN; 468 469 get_page(newpage); /* balloon reference */ 470 471 /* balloon's page migration 1st step -- inflate "newpage" */ 472 spin_lock_irqsave(&vb_dev_info->pages_lock, flags); 473 balloon_page_insert(vb_dev_info, newpage); 474 vb_dev_info->isolated_pages--; 475 __count_vm_event(BALLOON_MIGRATE); 476 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags); 477 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; 478 set_page_pfns(vb, vb->pfns, newpage); 479 tell_host(vb, vb->inflate_vq); 480 481 /* balloon's page migration 2nd step -- deflate "page" */ 482 balloon_page_delete(page); 483 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; 484 set_page_pfns(vb, vb->pfns, page); 485 tell_host(vb, vb->deflate_vq); 486 487 mutex_unlock(&vb->balloon_lock); 488 489 put_page(page); /* balloon reference */ 490 491 return MIGRATEPAGE_SUCCESS; 492 } 493 #endif /* CONFIG_BALLOON_COMPACTION */ 494 495 static int virtballoon_probe(struct virtio_device *vdev) 496 { 497 struct virtio_balloon *vb; 498 int err; 499 500 if (!vdev->config->get) { 501 dev_err(&vdev->dev, "%s failure: config access disabled\n", 502 __func__); 503 return -EINVAL; 504 } 505 506 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL); 507 if (!vb) { 508 err = -ENOMEM; 509 goto out; 510 } 511 512 INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func); 513 INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func); 514 spin_lock_init(&vb->stop_update_lock); 515 vb->stop_update = false; 516 vb->num_pages = 0; 517 mutex_init(&vb->balloon_lock); 518 init_waitqueue_head(&vb->acked); 519 vb->vdev = vdev; 520 521 balloon_devinfo_init(&vb->vb_dev_info); 522 #ifdef CONFIG_BALLOON_COMPACTION 523 vb->vb_dev_info.migratepage = virtballoon_migratepage; 524 #endif 525 526 err = init_vqs(vb); 527 if (err) 528 goto out_free_vb; 529 530 vb->nb.notifier_call = virtballoon_oom_notify; 531 vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY; 532 err = register_oom_notifier(&vb->nb); 533 if (err < 0) 534 goto out_oom_notify; 535 536 virtio_device_ready(vdev); 537 538 return 0; 539 540 out_oom_notify: 541 vdev->config->del_vqs(vdev); 542 out_free_vb: 543 kfree(vb); 544 out: 545 return err; 546 } 547 548 static void remove_common(struct virtio_balloon *vb) 549 { 550 /* There might be pages left in the balloon: free them. */ 551 while (vb->num_pages) 552 leak_balloon(vb, vb->num_pages); 553 update_balloon_size(vb); 554 555 /* Now we reset the device so we can clean up the queues. */ 556 vb->vdev->config->reset(vb->vdev); 557 558 vb->vdev->config->del_vqs(vb->vdev); 559 } 560 561 static void virtballoon_remove(struct virtio_device *vdev) 562 { 563 struct virtio_balloon *vb = vdev->priv; 564 565 unregister_oom_notifier(&vb->nb); 566 567 spin_lock_irq(&vb->stop_update_lock); 568 vb->stop_update = true; 569 spin_unlock_irq(&vb->stop_update_lock); 570 cancel_work_sync(&vb->update_balloon_size_work); 571 cancel_work_sync(&vb->update_balloon_stats_work); 572 573 remove_common(vb); 574 kfree(vb); 575 } 576 577 #ifdef CONFIG_PM_SLEEP 578 static int virtballoon_freeze(struct virtio_device *vdev) 579 { 580 struct virtio_balloon *vb = vdev->priv; 581 582 /* 583 * The workqueue is already frozen by the PM core before this 584 * function is called. 585 */ 586 remove_common(vb); 587 return 0; 588 } 589 590 static int virtballoon_restore(struct virtio_device *vdev) 591 { 592 struct virtio_balloon *vb = vdev->priv; 593 int ret; 594 595 ret = init_vqs(vdev->priv); 596 if (ret) 597 return ret; 598 599 virtio_device_ready(vdev); 600 601 if (towards_target(vb)) 602 virtballoon_changed(vdev); 603 update_balloon_size(vb); 604 return 0; 605 } 606 #endif 607 608 static unsigned int features[] = { 609 VIRTIO_BALLOON_F_MUST_TELL_HOST, 610 VIRTIO_BALLOON_F_STATS_VQ, 611 VIRTIO_BALLOON_F_DEFLATE_ON_OOM, 612 }; 613 614 static struct virtio_driver virtio_balloon_driver = { 615 .feature_table = features, 616 .feature_table_size = ARRAY_SIZE(features), 617 .driver.name = KBUILD_MODNAME, 618 .driver.owner = THIS_MODULE, 619 .id_table = id_table, 620 .probe = virtballoon_probe, 621 .remove = virtballoon_remove, 622 .config_changed = virtballoon_changed, 623 #ifdef CONFIG_PM_SLEEP 624 .freeze = virtballoon_freeze, 625 .restore = virtballoon_restore, 626 #endif 627 }; 628 629 module_virtio_driver(virtio_balloon_driver); 630 MODULE_DEVICE_TABLE(virtio, id_table); 631 MODULE_DESCRIPTION("Virtio balloon driver"); 632 MODULE_LICENSE("GPL"); 633