1// SPDX-License-Identifier: GPL-2.0-or-later 2/* 3 * Virtio balloon implementation, inspired by Dor Laor and Marcelo 4 * Tosatti's implementations. 5 * 6 * Copyright 2008 Rusty Russell IBM Corporation 7 */ 8 9#include <linux/virtio.h> 10#include <linux/virtio_balloon.h> 11#include <linux/swap.h> 12#include <linux/workqueue.h> 13#include <linux/delay.h> 14#include <linux/slab.h> 15#include <linux/module.h> 16#include <linux/balloon_compaction.h> 17#include <linux/oom.h> 18#include <linux/wait.h> 19#include <linux/mm.h> 20#include <linux/mount.h> 21#include <linux/magic.h> 22#include <linux/pseudo_fs.h> 23#include <linux/page_reporting.h> 24 25/* 26 * Balloon device works in 4K page units. So each page is pointed to by 27 * multiple balloon pages. All memory counters in this driver are in balloon 28 * page units. 29 */ 30#define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT) 31#define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256 32/* Maximum number of (4k) pages to deflate on OOM notifications. */ 33#define VIRTIO_BALLOON_OOM_NR_PAGES 256 34#define VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY 80 35 36#define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \ 37 __GFP_NOMEMALLOC) 38/* The order of free page blocks to report to host */ 39#define VIRTIO_BALLOON_HINT_BLOCK_ORDER (MAX_ORDER - 1) 40/* The size of a free page block in bytes */ 41#define VIRTIO_BALLOON_HINT_BLOCK_BYTES \ 42 (1 << (VIRTIO_BALLOON_HINT_BLOCK_ORDER + PAGE_SHIFT)) 43#define VIRTIO_BALLOON_HINT_BLOCK_PAGES (1 << VIRTIO_BALLOON_HINT_BLOCK_ORDER) 44 45#ifdef CONFIG_BALLOON_COMPACTION 46static struct vfsmount *balloon_mnt; 47#endif 48 49enum virtio_balloon_vq { 50 VIRTIO_BALLOON_VQ_INFLATE, 51 VIRTIO_BALLOON_VQ_DEFLATE, 52 VIRTIO_BALLOON_VQ_STATS, 53 VIRTIO_BALLOON_VQ_FREE_PAGE, 54 VIRTIO_BALLOON_VQ_REPORTING, 55 VIRTIO_BALLOON_VQ_MAX 56}; 57 58enum virtio_balloon_config_read { 59 VIRTIO_BALLOON_CONFIG_READ_CMD_ID = 0, 60}; 61 62struct virtio_balloon { 63 struct virtio_device *vdev; 64 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq; 65 66 /* Balloon's own wq for cpu-intensive work items */ 67 struct workqueue_struct *balloon_wq; 68 /* The free page reporting work item submitted to the balloon wq */ 69 struct work_struct report_free_page_work; 70 71 /* The balloon servicing is delegated to a freezable workqueue. */ 72 struct work_struct update_balloon_stats_work; 73 struct work_struct update_balloon_size_work; 74 75 /* Prevent updating balloon when it is being canceled. */ 76 spinlock_t stop_update_lock; 77 bool stop_update; 78 /* Bitmap to indicate if reading the related config fields are needed */ 79 unsigned long config_read_bitmap; 80 81 /* The list of allocated free pages, waiting to be given back to mm */ 82 struct list_head free_page_list; 83 spinlock_t free_page_list_lock; 84 /* The number of free page blocks on the above list */ 85 unsigned long num_free_page_blocks; 86 /* 87 * The cmd id received from host. 88 * Read it via virtio_balloon_cmd_id_received to get the latest value 89 * sent from host. 90 */ 91 u32 cmd_id_received_cache; 92 /* The cmd id that is actively in use */ 93 __virtio32 cmd_id_active; 94 /* Buffer to store the stop sign */ 95 __virtio32 cmd_id_stop; 96 97 /* Waiting for host to ack the pages we released. */ 98 wait_queue_head_t acked; 99 100 /* Number of balloon pages we've told the Host we're not using. */ 101 unsigned int num_pages; 102 /* 103 * The pages we've told the Host we're not using are enqueued 104 * at vb_dev_info->pages list. 105 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE 106 * to num_pages above. 107 */ 108 struct balloon_dev_info vb_dev_info; 109 110 /* Synchronize access/update to this struct virtio_balloon elements */ 111 struct mutex balloon_lock; 112 113 /* The array of pfns we tell the Host about. */ 114 unsigned int num_pfns; 115 __virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX]; 116 117 /* Memory statistics */ 118 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR]; 119 120 /* Shrinker to return free pages - VIRTIO_BALLOON_F_FREE_PAGE_HINT */ 121 struct shrinker shrinker; 122 123 /* OOM notifier to deflate on OOM - VIRTIO_BALLOON_F_DEFLATE_ON_OOM */ 124 struct notifier_block oom_nb; 125 126 /* Free page reporting device */ 127 struct virtqueue *reporting_vq; 128 struct page_reporting_dev_info pr_dev_info; 129}; 130 131static const struct virtio_device_id id_table[] = { 132 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID }, 133 { 0 }, 134}; 135 136static u32 page_to_balloon_pfn(struct page *page) 137{ 138 unsigned long pfn = page_to_pfn(page); 139 140 BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT); 141 /* Convert pfn from Linux page size to balloon page size. */ 142 return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE; 143} 144 145static void balloon_ack(struct virtqueue *vq) 146{ 147 struct virtio_balloon *vb = vq->vdev->priv; 148 149 wake_up(&vb->acked); 150} 151 152static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq) 153{ 154 struct scatterlist sg; 155 unsigned int len; 156 157 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns); 158 159 /* We should always be able to add one buffer to an empty queue. */ 160 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); 161 virtqueue_kick(vq); 162 163 /* When host has read buffer, this completes via balloon_ack */ 164 wait_event(vb->acked, virtqueue_get_buf(vq, &len)); 165 166} 167 168static int virtballoon_free_page_report(struct page_reporting_dev_info *pr_dev_info, 169 struct scatterlist *sg, unsigned int nents) 170{ 171 struct virtio_balloon *vb = 172 container_of(pr_dev_info, struct virtio_balloon, pr_dev_info); 173 struct virtqueue *vq = vb->reporting_vq; 174 unsigned int unused, err; 175 176 /* We should always be able to add these buffers to an empty queue. */ 177 err = virtqueue_add_inbuf(vq, sg, nents, vb, GFP_NOWAIT | __GFP_NOWARN); 178 179 /* 180 * In the extremely unlikely case that something has occurred and we 181 * are able to trigger an error we will simply display a warning 182 * and exit without actually processing the pages. 183 */ 184 if (WARN_ON_ONCE(err)) 185 return err; 186 187 virtqueue_kick(vq); 188 189 /* When host has read buffer, this completes via balloon_ack */ 190 wait_event(vb->acked, virtqueue_get_buf(vq, &unused)); 191 192 return 0; 193} 194 195static void set_page_pfns(struct virtio_balloon *vb, 196 __virtio32 pfns[], struct page *page) 197{ 198 unsigned int i; 199 200 BUILD_BUG_ON(VIRTIO_BALLOON_PAGES_PER_PAGE > VIRTIO_BALLOON_ARRAY_PFNS_MAX); 201 202 /* 203 * Set balloon pfns pointing at this page. 204 * Note that the first pfn points at start of the page. 205 */ 206 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++) 207 pfns[i] = cpu_to_virtio32(vb->vdev, 208 page_to_balloon_pfn(page) + i); 209} 210 211static unsigned fill_balloon(struct virtio_balloon *vb, size_t num) 212{ 213 unsigned num_allocated_pages; 214 unsigned num_pfns; 215 struct page *page; 216 LIST_HEAD(pages); 217 218 /* We can only do one array worth at a time. */ 219 num = min(num, ARRAY_SIZE(vb->pfns)); 220 221 for (num_pfns = 0; num_pfns < num; 222 num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { 223 struct page *page = balloon_page_alloc(); 224 225 if (!page) { 226 dev_info_ratelimited(&vb->vdev->dev, 227 "Out of puff! Can't get %u pages\n", 228 VIRTIO_BALLOON_PAGES_PER_PAGE); 229 /* Sleep for at least 1/5 of a second before retry. */ 230 msleep(200); 231 break; 232 } 233 234 balloon_page_push(&pages, page); 235 } 236 237 mutex_lock(&vb->balloon_lock); 238 239 vb->num_pfns = 0; 240 241 while ((page = balloon_page_pop(&pages))) { 242 balloon_page_enqueue(&vb->vb_dev_info, page); 243 244 set_page_pfns(vb, vb->pfns + vb->num_pfns, page); 245 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE; 246 if (!virtio_has_feature(vb->vdev, 247 VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 248 adjust_managed_page_count(page, -1); 249 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE; 250 } 251 252 num_allocated_pages = vb->num_pfns; 253 /* Did we get any? */ 254 if (vb->num_pfns != 0) 255 tell_host(vb, vb->inflate_vq); 256 mutex_unlock(&vb->balloon_lock); 257 258 return num_allocated_pages; 259} 260 261static void release_pages_balloon(struct virtio_balloon *vb, 262 struct list_head *pages) 263{ 264 struct page *page, *next; 265 266 list_for_each_entry_safe(page, next, pages, lru) { 267 if (!virtio_has_feature(vb->vdev, 268 VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 269 adjust_managed_page_count(page, 1); 270 list_del(&page->lru); 271 put_page(page); /* balloon reference */ 272 } 273} 274 275static unsigned leak_balloon(struct virtio_balloon *vb, size_t num) 276{ 277 unsigned num_freed_pages; 278 struct page *page; 279 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info; 280 LIST_HEAD(pages); 281 282 /* We can only do one array worth at a time. */ 283 num = min(num, ARRAY_SIZE(vb->pfns)); 284 285 mutex_lock(&vb->balloon_lock); 286 /* We can't release more pages than taken */ 287 num = min(num, (size_t)vb->num_pages); 288 for (vb->num_pfns = 0; vb->num_pfns < num; 289 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { 290 page = balloon_page_dequeue(vb_dev_info); 291 if (!page) 292 break; 293 set_page_pfns(vb, vb->pfns + vb->num_pfns, page); 294 list_add(&page->lru, &pages); 295 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE; 296 } 297 298 num_freed_pages = vb->num_pfns; 299 /* 300 * Note that if 301 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST); 302 * is true, we *have* to do it in this order 303 */ 304 if (vb->num_pfns != 0) 305 tell_host(vb, vb->deflate_vq); 306 release_pages_balloon(vb, &pages); 307 mutex_unlock(&vb->balloon_lock); 308 return num_freed_pages; 309} 310 311static inline void update_stat(struct virtio_balloon *vb, int idx, 312 u16 tag, u64 val) 313{ 314 BUG_ON(idx >= VIRTIO_BALLOON_S_NR); 315 vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag); 316 vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val); 317} 318 319#define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT) 320 321static unsigned int update_balloon_stats(struct virtio_balloon *vb) 322{ 323 unsigned long events[NR_VM_EVENT_ITEMS]; 324 struct sysinfo i; 325 unsigned int idx = 0; 326 long available; 327 unsigned long caches; 328 329 all_vm_events(events); 330 si_meminfo(&i); 331 332 available = si_mem_available(); 333 caches = global_node_page_state(NR_FILE_PAGES); 334 335#ifdef CONFIG_VM_EVENT_COUNTERS 336 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN, 337 pages_to_bytes(events[PSWPIN])); 338 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT, 339 pages_to_bytes(events[PSWPOUT])); 340 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]); 341 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]); 342#ifdef CONFIG_HUGETLB_PAGE 343 update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC, 344 events[HTLB_BUDDY_PGALLOC]); 345 update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL, 346 events[HTLB_BUDDY_PGALLOC_FAIL]); 347#endif 348#endif 349 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE, 350 pages_to_bytes(i.freeram)); 351 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT, 352 pages_to_bytes(i.totalram)); 353 update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL, 354 pages_to_bytes(available)); 355 update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES, 356 pages_to_bytes(caches)); 357 358 return idx; 359} 360 361/* 362 * While most virtqueues communicate guest-initiated requests to the hypervisor, 363 * the stats queue operates in reverse. The driver initializes the virtqueue 364 * with a single buffer. From that point forward, all conversations consist of 365 * a hypervisor request (a call to this function) which directs us to refill 366 * the virtqueue with a fresh stats buffer. Since stats collection can sleep, 367 * we delegate the job to a freezable workqueue that will do the actual work via 368 * stats_handle_request(). 369 */ 370static void stats_request(struct virtqueue *vq) 371{ 372 struct virtio_balloon *vb = vq->vdev->priv; 373 374 spin_lock(&vb->stop_update_lock); 375 if (!vb->stop_update) 376 queue_work(system_freezable_wq, &vb->update_balloon_stats_work); 377 spin_unlock(&vb->stop_update_lock); 378} 379 380static void stats_handle_request(struct virtio_balloon *vb) 381{ 382 struct virtqueue *vq; 383 struct scatterlist sg; 384 unsigned int len, num_stats; 385 386 num_stats = update_balloon_stats(vb); 387 388 vq = vb->stats_vq; 389 if (!virtqueue_get_buf(vq, &len)) 390 return; 391 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats); 392 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); 393 virtqueue_kick(vq); 394} 395 396static inline s64 towards_target(struct virtio_balloon *vb) 397{ 398 s64 target; 399 u32 num_pages; 400 401 /* Legacy balloon config space is LE, unlike all other devices. */ 402 virtio_cread_le(vb->vdev, struct virtio_balloon_config, num_pages, 403 &num_pages); 404 405 /* 406 * Aligned up to guest page size to avoid inflating and deflating 407 * balloon endlessly. 408 */ 409 target = ALIGN(num_pages, VIRTIO_BALLOON_PAGES_PER_PAGE); 410 return target - vb->num_pages; 411} 412 413/* Gives back @num_to_return blocks of free pages to mm. */ 414static unsigned long return_free_pages_to_mm(struct virtio_balloon *vb, 415 unsigned long num_to_return) 416{ 417 struct page *page; 418 unsigned long num_returned; 419 420 spin_lock_irq(&vb->free_page_list_lock); 421 for (num_returned = 0; num_returned < num_to_return; num_returned++) { 422 page = balloon_page_pop(&vb->free_page_list); 423 if (!page) 424 break; 425 free_pages((unsigned long)page_address(page), 426 VIRTIO_BALLOON_HINT_BLOCK_ORDER); 427 } 428 vb->num_free_page_blocks -= num_returned; 429 spin_unlock_irq(&vb->free_page_list_lock); 430 431 return num_returned; 432} 433 434static void virtio_balloon_queue_free_page_work(struct virtio_balloon *vb) 435{ 436 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 437 return; 438 439 /* No need to queue the work if the bit was already set. */ 440 if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID, 441 &vb->config_read_bitmap)) 442 return; 443 444 queue_work(vb->balloon_wq, &vb->report_free_page_work); 445} 446 447static void virtballoon_changed(struct virtio_device *vdev) 448{ 449 struct virtio_balloon *vb = vdev->priv; 450 unsigned long flags; 451 452 spin_lock_irqsave(&vb->stop_update_lock, flags); 453 if (!vb->stop_update) { 454 queue_work(system_freezable_wq, 455 &vb->update_balloon_size_work); 456 virtio_balloon_queue_free_page_work(vb); 457 } 458 spin_unlock_irqrestore(&vb->stop_update_lock, flags); 459} 460 461static void update_balloon_size(struct virtio_balloon *vb) 462{ 463 u32 actual = vb->num_pages; 464 465 /* Legacy balloon config space is LE, unlike all other devices. */ 466 virtio_cwrite_le(vb->vdev, struct virtio_balloon_config, actual, 467 &actual); 468} 469 470static void update_balloon_stats_func(struct work_struct *work) 471{ 472 struct virtio_balloon *vb; 473 474 vb = container_of(work, struct virtio_balloon, 475 update_balloon_stats_work); 476 stats_handle_request(vb); 477} 478 479static void update_balloon_size_func(struct work_struct *work) 480{ 481 struct virtio_balloon *vb; 482 s64 diff; 483 484 vb = container_of(work, struct virtio_balloon, 485 update_balloon_size_work); 486 diff = towards_target(vb); 487 488 if (!diff) 489 return; 490 491 if (diff > 0) 492 diff -= fill_balloon(vb, diff); 493 else 494 diff += leak_balloon(vb, -diff); 495 update_balloon_size(vb); 496 497 if (diff) 498 queue_work(system_freezable_wq, work); 499} 500 501static int init_vqs(struct virtio_balloon *vb) 502{ 503 struct virtqueue *vqs[VIRTIO_BALLOON_VQ_MAX]; 504 vq_callback_t *callbacks[VIRTIO_BALLOON_VQ_MAX]; 505 const char *names[VIRTIO_BALLOON_VQ_MAX]; 506 int err; 507 508 /* 509 * Inflateq and deflateq are used unconditionally. The names[] 510 * will be NULL if the related feature is not enabled, which will 511 * cause no allocation for the corresponding virtqueue in find_vqs. 512 */ 513 callbacks[VIRTIO_BALLOON_VQ_INFLATE] = balloon_ack; 514 names[VIRTIO_BALLOON_VQ_INFLATE] = "inflate"; 515 callbacks[VIRTIO_BALLOON_VQ_DEFLATE] = balloon_ack; 516 names[VIRTIO_BALLOON_VQ_DEFLATE] = "deflate"; 517 callbacks[VIRTIO_BALLOON_VQ_STATS] = NULL; 518 names[VIRTIO_BALLOON_VQ_STATS] = NULL; 519 callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL; 520 names[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL; 521 names[VIRTIO_BALLOON_VQ_REPORTING] = NULL; 522 523 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { 524 names[VIRTIO_BALLOON_VQ_STATS] = "stats"; 525 callbacks[VIRTIO_BALLOON_VQ_STATS] = stats_request; 526 } 527 528 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { 529 names[VIRTIO_BALLOON_VQ_FREE_PAGE] = "free_page_vq"; 530 callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL; 531 } 532 533 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) { 534 names[VIRTIO_BALLOON_VQ_REPORTING] = "reporting_vq"; 535 callbacks[VIRTIO_BALLOON_VQ_REPORTING] = balloon_ack; 536 } 537 538 err = vb->vdev->config->find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX, 539 vqs, callbacks, names, NULL, NULL); 540 if (err) 541 return err; 542 543 vb->inflate_vq = vqs[VIRTIO_BALLOON_VQ_INFLATE]; 544 vb->deflate_vq = vqs[VIRTIO_BALLOON_VQ_DEFLATE]; 545 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { 546 struct scatterlist sg; 547 unsigned int num_stats; 548 vb->stats_vq = vqs[VIRTIO_BALLOON_VQ_STATS]; 549 550 /* 551 * Prime this virtqueue with one buffer so the hypervisor can 552 * use it to signal us later (it can't be broken yet!). 553 */ 554 num_stats = update_balloon_stats(vb); 555 556 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats); 557 err = virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, 558 GFP_KERNEL); 559 if (err) { 560 dev_warn(&vb->vdev->dev, "%s: add stat_vq failed\n", 561 __func__); 562 return err; 563 } 564 virtqueue_kick(vb->stats_vq); 565 } 566 567 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 568 vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE]; 569 570 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) 571 vb->reporting_vq = vqs[VIRTIO_BALLOON_VQ_REPORTING]; 572 573 return 0; 574} 575 576static u32 virtio_balloon_cmd_id_received(struct virtio_balloon *vb) 577{ 578 if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID, 579 &vb->config_read_bitmap)) { 580 /* Legacy balloon config space is LE, unlike all other devices. */ 581 virtio_cread_le(vb->vdev, struct virtio_balloon_config, 582 free_page_hint_cmd_id, 583 &vb->cmd_id_received_cache); 584 } 585 586 return vb->cmd_id_received_cache; 587} 588 589static int send_cmd_id_start(struct virtio_balloon *vb) 590{ 591 struct scatterlist sg; 592 struct virtqueue *vq = vb->free_page_vq; 593 int err, unused; 594 595 /* Detach all the used buffers from the vq */ 596 while (virtqueue_get_buf(vq, &unused)) 597 ; 598 599 vb->cmd_id_active = cpu_to_virtio32(vb->vdev, 600 virtio_balloon_cmd_id_received(vb)); 601 sg_init_one(&sg, &vb->cmd_id_active, sizeof(vb->cmd_id_active)); 602 err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_active, GFP_KERNEL); 603 if (!err) 604 virtqueue_kick(vq); 605 return err; 606} 607 608static int send_cmd_id_stop(struct virtio_balloon *vb) 609{ 610 struct scatterlist sg; 611 struct virtqueue *vq = vb->free_page_vq; 612 int err, unused; 613 614 /* Detach all the used buffers from the vq */ 615 while (virtqueue_get_buf(vq, &unused)) 616 ; 617 618 sg_init_one(&sg, &vb->cmd_id_stop, sizeof(vb->cmd_id_stop)); 619 err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_stop, GFP_KERNEL); 620 if (!err) 621 virtqueue_kick(vq); 622 return err; 623} 624 625static int get_free_page_and_send(struct virtio_balloon *vb) 626{ 627 struct virtqueue *vq = vb->free_page_vq; 628 struct page *page; 629 struct scatterlist sg; 630 int err, unused; 631 void *p; 632 633 /* Detach all the used buffers from the vq */ 634 while (virtqueue_get_buf(vq, &unused)) 635 ; 636 637 page = alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG, 638 VIRTIO_BALLOON_HINT_BLOCK_ORDER); 639 /* 640 * When the allocation returns NULL, it indicates that we have got all 641 * the possible free pages, so return -EINTR to stop. 642 */ 643 if (!page) 644 return -EINTR; 645 646 p = page_address(page); 647 sg_init_one(&sg, p, VIRTIO_BALLOON_HINT_BLOCK_BYTES); 648 /* There is always 1 entry reserved for the cmd id to use. */ 649 if (vq->num_free > 1) { 650 err = virtqueue_add_inbuf(vq, &sg, 1, p, GFP_KERNEL); 651 if (unlikely(err)) { 652 free_pages((unsigned long)p, 653 VIRTIO_BALLOON_HINT_BLOCK_ORDER); 654 return err; 655 } 656 virtqueue_kick(vq); 657 spin_lock_irq(&vb->free_page_list_lock); 658 balloon_page_push(&vb->free_page_list, page); 659 vb->num_free_page_blocks++; 660 spin_unlock_irq(&vb->free_page_list_lock); 661 } else { 662 /* 663 * The vq has no available entry to add this page block, so 664 * just free it. 665 */ 666 free_pages((unsigned long)p, VIRTIO_BALLOON_HINT_BLOCK_ORDER); 667 } 668 669 return 0; 670} 671 672static int send_free_pages(struct virtio_balloon *vb) 673{ 674 int err; 675 u32 cmd_id_active; 676 677 while (1) { 678 /* 679 * If a stop id or a new cmd id was just received from host, 680 * stop the reporting. 681 */ 682 cmd_id_active = virtio32_to_cpu(vb->vdev, vb->cmd_id_active); 683 if (unlikely(cmd_id_active != 684 virtio_balloon_cmd_id_received(vb))) 685 break; 686 687 /* 688 * The free page blocks are allocated and sent to host one by 689 * one. 690 */ 691 err = get_free_page_and_send(vb); 692 if (err == -EINTR) 693 break; 694 else if (unlikely(err)) 695 return err; 696 } 697 698 return 0; 699} 700 701static void virtio_balloon_report_free_page(struct virtio_balloon *vb) 702{ 703 int err; 704 struct device *dev = &vb->vdev->dev; 705 706 /* Start by sending the received cmd id to host with an outbuf. */ 707 err = send_cmd_id_start(vb); 708 if (unlikely(err)) 709 dev_err(dev, "Failed to send a start id, err = %d\n", err); 710 711 err = send_free_pages(vb); 712 if (unlikely(err)) 713 dev_err(dev, "Failed to send a free page, err = %d\n", err); 714 715 /* End by sending a stop id to host with an outbuf. */ 716 err = send_cmd_id_stop(vb); 717 if (unlikely(err)) 718 dev_err(dev, "Failed to send a stop id, err = %d\n", err); 719} 720 721static void report_free_page_func(struct work_struct *work) 722{ 723 struct virtio_balloon *vb = container_of(work, struct virtio_balloon, 724 report_free_page_work); 725 u32 cmd_id_received; 726 727 cmd_id_received = virtio_balloon_cmd_id_received(vb); 728 if (cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) { 729 /* Pass ULONG_MAX to give back all the free pages */ 730 return_free_pages_to_mm(vb, ULONG_MAX); 731 } else if (cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP && 732 cmd_id_received != 733 virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) { 734 virtio_balloon_report_free_page(vb); 735 } 736} 737 738#ifdef CONFIG_BALLOON_COMPACTION 739/* 740 * virtballoon_migratepage - perform the balloon page migration on behalf of 741 * a compation thread. (called under page lock) 742 * @vb_dev_info: the balloon device 743 * @newpage: page that will replace the isolated page after migration finishes. 744 * @page : the isolated (old) page that is about to be migrated to newpage. 745 * @mode : compaction mode -- not used for balloon page migration. 746 * 747 * After a ballooned page gets isolated by compaction procedures, this is the 748 * function that performs the page migration on behalf of a compaction thread 749 * The page migration for virtio balloon is done in a simple swap fashion which 750 * follows these two macro steps: 751 * 1) insert newpage into vb->pages list and update the host about it; 752 * 2) update the host about the old page removed from vb->pages list; 753 * 754 * This function preforms the balloon page migration task. 755 * Called through balloon_mapping->a_ops->migratepage 756 */ 757static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info, 758 struct page *newpage, struct page *page, enum migrate_mode mode) 759{ 760 struct virtio_balloon *vb = container_of(vb_dev_info, 761 struct virtio_balloon, vb_dev_info); 762 unsigned long flags; 763 764 /* 765 * In order to avoid lock contention while migrating pages concurrently 766 * to leak_balloon() or fill_balloon() we just give up the balloon_lock 767 * this turn, as it is easier to retry the page migration later. 768 * This also prevents fill_balloon() getting stuck into a mutex 769 * recursion in the case it ends up triggering memory compaction 770 * while it is attempting to inflate the ballon. 771 */ 772 if (!mutex_trylock(&vb->balloon_lock)) 773 return -EAGAIN; 774 775 get_page(newpage); /* balloon reference */ 776 777 /* 778 * When we migrate a page to a different zone and adjusted the 779 * managed page count when inflating, we have to fixup the count of 780 * both involved zones. 781 */ 782 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM) && 783 page_zone(page) != page_zone(newpage)) { 784 adjust_managed_page_count(page, 1); 785 adjust_managed_page_count(newpage, -1); 786 } 787 788 /* balloon's page migration 1st step -- inflate "newpage" */ 789 spin_lock_irqsave(&vb_dev_info->pages_lock, flags); 790 balloon_page_insert(vb_dev_info, newpage); 791 vb_dev_info->isolated_pages--; 792 __count_vm_event(BALLOON_MIGRATE); 793 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags); 794 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; 795 set_page_pfns(vb, vb->pfns, newpage); 796 tell_host(vb, vb->inflate_vq); 797 798 /* balloon's page migration 2nd step -- deflate "page" */ 799 spin_lock_irqsave(&vb_dev_info->pages_lock, flags); 800 balloon_page_delete(page); 801 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags); 802 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; 803 set_page_pfns(vb, vb->pfns, page); 804 tell_host(vb, vb->deflate_vq); 805 806 mutex_unlock(&vb->balloon_lock); 807 808 put_page(page); /* balloon reference */ 809 810 return MIGRATEPAGE_SUCCESS; 811} 812 813static int balloon_init_fs_context(struct fs_context *fc) 814{ 815 return init_pseudo(fc, BALLOON_KVM_MAGIC) ? 0 : -ENOMEM; 816} 817 818static struct file_system_type balloon_fs = { 819 .name = "balloon-kvm", 820 .init_fs_context = balloon_init_fs_context, 821 .kill_sb = kill_anon_super, 822}; 823 824#endif /* CONFIG_BALLOON_COMPACTION */ 825 826static unsigned long shrink_free_pages(struct virtio_balloon *vb, 827 unsigned long pages_to_free) 828{ 829 unsigned long blocks_to_free, blocks_freed; 830 831 pages_to_free = round_up(pages_to_free, 832 VIRTIO_BALLOON_HINT_BLOCK_PAGES); 833 blocks_to_free = pages_to_free / VIRTIO_BALLOON_HINT_BLOCK_PAGES; 834 blocks_freed = return_free_pages_to_mm(vb, blocks_to_free); 835 836 return blocks_freed * VIRTIO_BALLOON_HINT_BLOCK_PAGES; 837} 838 839static unsigned long virtio_balloon_shrinker_scan(struct shrinker *shrinker, 840 struct shrink_control *sc) 841{ 842 struct virtio_balloon *vb = container_of(shrinker, 843 struct virtio_balloon, shrinker); 844 845 return shrink_free_pages(vb, sc->nr_to_scan); 846} 847 848static unsigned long virtio_balloon_shrinker_count(struct shrinker *shrinker, 849 struct shrink_control *sc) 850{ 851 struct virtio_balloon *vb = container_of(shrinker, 852 struct virtio_balloon, shrinker); 853 854 return vb->num_free_page_blocks * VIRTIO_BALLOON_HINT_BLOCK_PAGES; 855} 856 857static int virtio_balloon_oom_notify(struct notifier_block *nb, 858 unsigned long dummy, void *parm) 859{ 860 struct virtio_balloon *vb = container_of(nb, 861 struct virtio_balloon, oom_nb); 862 unsigned long *freed = parm; 863 864 *freed += leak_balloon(vb, VIRTIO_BALLOON_OOM_NR_PAGES) / 865 VIRTIO_BALLOON_PAGES_PER_PAGE; 866 update_balloon_size(vb); 867 868 return NOTIFY_OK; 869} 870 871static void virtio_balloon_unregister_shrinker(struct virtio_balloon *vb) 872{ 873 unregister_shrinker(&vb->shrinker); 874} 875 876static int virtio_balloon_register_shrinker(struct virtio_balloon *vb) 877{ 878 vb->shrinker.scan_objects = virtio_balloon_shrinker_scan; 879 vb->shrinker.count_objects = virtio_balloon_shrinker_count; 880 vb->shrinker.seeks = DEFAULT_SEEKS; 881 882 return register_shrinker(&vb->shrinker); 883} 884 885static int virtballoon_probe(struct virtio_device *vdev) 886{ 887 struct virtio_balloon *vb; 888 int err; 889 890 if (!vdev->config->get) { 891 dev_err(&vdev->dev, "%s failure: config access disabled\n", 892 __func__); 893 return -EINVAL; 894 } 895 896 vdev->priv = vb = kzalloc(sizeof(*vb), GFP_KERNEL); 897 if (!vb) { 898 err = -ENOMEM; 899 goto out; 900 } 901 902 INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func); 903 INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func); 904 spin_lock_init(&vb->stop_update_lock); 905 mutex_init(&vb->balloon_lock); 906 init_waitqueue_head(&vb->acked); 907 vb->vdev = vdev; 908 909 balloon_devinfo_init(&vb->vb_dev_info); 910 911 err = init_vqs(vb); 912 if (err) 913 goto out_free_vb; 914 915#ifdef CONFIG_BALLOON_COMPACTION 916 balloon_mnt = kern_mount(&balloon_fs); 917 if (IS_ERR(balloon_mnt)) { 918 err = PTR_ERR(balloon_mnt); 919 goto out_del_vqs; 920 } 921 922 vb->vb_dev_info.migratepage = virtballoon_migratepage; 923 vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb); 924 if (IS_ERR(vb->vb_dev_info.inode)) { 925 err = PTR_ERR(vb->vb_dev_info.inode); 926 goto out_kern_unmount; 927 } 928 vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops; 929#endif 930 if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { 931 /* 932 * There is always one entry reserved for cmd id, so the ring 933 * size needs to be at least two to report free page hints. 934 */ 935 if (virtqueue_get_vring_size(vb->free_page_vq) < 2) { 936 err = -ENOSPC; 937 goto out_iput; 938 } 939 vb->balloon_wq = alloc_workqueue("balloon-wq", 940 WQ_FREEZABLE | WQ_CPU_INTENSIVE, 0); 941 if (!vb->balloon_wq) { 942 err = -ENOMEM; 943 goto out_iput; 944 } 945 INIT_WORK(&vb->report_free_page_work, report_free_page_func); 946 vb->cmd_id_received_cache = VIRTIO_BALLOON_CMD_ID_STOP; 947 vb->cmd_id_active = cpu_to_virtio32(vb->vdev, 948 VIRTIO_BALLOON_CMD_ID_STOP); 949 vb->cmd_id_stop = cpu_to_virtio32(vb->vdev, 950 VIRTIO_BALLOON_CMD_ID_STOP); 951 spin_lock_init(&vb->free_page_list_lock); 952 INIT_LIST_HEAD(&vb->free_page_list); 953 /* 954 * We're allowed to reuse any free pages, even if they are 955 * still to be processed by the host. 956 */ 957 err = virtio_balloon_register_shrinker(vb); 958 if (err) 959 goto out_del_balloon_wq; 960 } 961 962 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) { 963 vb->oom_nb.notifier_call = virtio_balloon_oom_notify; 964 vb->oom_nb.priority = VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY; 965 err = register_oom_notifier(&vb->oom_nb); 966 if (err < 0) 967 goto out_unregister_shrinker; 968 } 969 970 if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) { 971 /* Start with poison val of 0 representing general init */ 972 __u32 poison_val = 0; 973 974 /* 975 * Let the hypervisor know that we are expecting a 976 * specific value to be written back in balloon pages. 977 * 978 * If the PAGE_POISON value was larger than a byte we would 979 * need to byte swap poison_val here to guarantee it is 980 * little-endian. However for now it is a single byte so we 981 * can pass it as-is. 982 */ 983 if (!want_init_on_free()) 984 memset(&poison_val, PAGE_POISON, sizeof(poison_val)); 985 986 virtio_cwrite_le(vb->vdev, struct virtio_balloon_config, 987 poison_val, &poison_val); 988 } 989 990 vb->pr_dev_info.report = virtballoon_free_page_report; 991 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) { 992 unsigned int capacity; 993 994 capacity = virtqueue_get_vring_size(vb->reporting_vq); 995 if (capacity < PAGE_REPORTING_CAPACITY) { 996 err = -ENOSPC; 997 goto out_unregister_oom; 998 } 999 1000 err = page_reporting_register(&vb->pr_dev_info); 1001 if (err) 1002 goto out_unregister_oom; 1003 } 1004 1005 virtio_device_ready(vdev); 1006 1007 if (towards_target(vb)) 1008 virtballoon_changed(vdev); 1009 return 0; 1010 1011out_unregister_oom: 1012 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 1013 unregister_oom_notifier(&vb->oom_nb); 1014out_unregister_shrinker: 1015 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 1016 virtio_balloon_unregister_shrinker(vb); 1017out_del_balloon_wq: 1018 if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 1019 destroy_workqueue(vb->balloon_wq); 1020out_iput: 1021#ifdef CONFIG_BALLOON_COMPACTION 1022 iput(vb->vb_dev_info.inode); 1023out_kern_unmount: 1024 kern_unmount(balloon_mnt); 1025out_del_vqs: 1026#endif 1027 vdev->config->del_vqs(vdev); 1028out_free_vb: 1029 kfree(vb); 1030out: 1031 return err; 1032} 1033 1034static void remove_common(struct virtio_balloon *vb) 1035{ 1036 /* There might be pages left in the balloon: free them. */ 1037 while (vb->num_pages) 1038 leak_balloon(vb, vb->num_pages); 1039 update_balloon_size(vb); 1040 1041 /* There might be free pages that are being reported: release them. */ 1042 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 1043 return_free_pages_to_mm(vb, ULONG_MAX); 1044 1045 /* Now we reset the device so we can clean up the queues. */ 1046 vb->vdev->config->reset(vb->vdev); 1047 1048 vb->vdev->config->del_vqs(vb->vdev); 1049} 1050 1051static void virtballoon_remove(struct virtio_device *vdev) 1052{ 1053 struct virtio_balloon *vb = vdev->priv; 1054 1055 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) 1056 page_reporting_unregister(&vb->pr_dev_info); 1057 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 1058 unregister_oom_notifier(&vb->oom_nb); 1059 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 1060 virtio_balloon_unregister_shrinker(vb); 1061 spin_lock_irq(&vb->stop_update_lock); 1062 vb->stop_update = true; 1063 spin_unlock_irq(&vb->stop_update_lock); 1064 cancel_work_sync(&vb->update_balloon_size_work); 1065 cancel_work_sync(&vb->update_balloon_stats_work); 1066 1067 if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { 1068 cancel_work_sync(&vb->report_free_page_work); 1069 destroy_workqueue(vb->balloon_wq); 1070 } 1071 1072 remove_common(vb); 1073#ifdef CONFIG_BALLOON_COMPACTION 1074 if (vb->vb_dev_info.inode) 1075 iput(vb->vb_dev_info.inode); 1076 1077 kern_unmount(balloon_mnt); 1078#endif 1079 kfree(vb); 1080} 1081 1082#ifdef CONFIG_PM_SLEEP 1083static int virtballoon_freeze(struct virtio_device *vdev) 1084{ 1085 struct virtio_balloon *vb = vdev->priv; 1086 1087 /* 1088 * The workqueue is already frozen by the PM core before this 1089 * function is called. 1090 */ 1091 remove_common(vb); 1092 return 0; 1093} 1094 1095static int virtballoon_restore(struct virtio_device *vdev) 1096{ 1097 struct virtio_balloon *vb = vdev->priv; 1098 int ret; 1099 1100 ret = init_vqs(vdev->priv); 1101 if (ret) 1102 return ret; 1103 1104 virtio_device_ready(vdev); 1105 1106 if (towards_target(vb)) 1107 virtballoon_changed(vdev); 1108 update_balloon_size(vb); 1109 return 0; 1110} 1111#endif 1112 1113static int virtballoon_validate(struct virtio_device *vdev) 1114{ 1115 /* 1116 * Inform the hypervisor that our pages are poisoned or 1117 * initialized. If we cannot do that then we should disable 1118 * page reporting as it could potentially change the contents 1119 * of our free pages. 1120 */ 1121 if (!want_init_on_free() && 1122 (IS_ENABLED(CONFIG_PAGE_POISONING_NO_SANITY) || 1123 !page_poisoning_enabled())) 1124 __virtio_clear_bit(vdev, VIRTIO_BALLOON_F_PAGE_POISON); 1125 else if (!virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) 1126 __virtio_clear_bit(vdev, VIRTIO_BALLOON_F_REPORTING); 1127 1128 __virtio_clear_bit(vdev, VIRTIO_F_ACCESS_PLATFORM); 1129 return 0; 1130} 1131 1132static unsigned int features[] = { 1133 VIRTIO_BALLOON_F_MUST_TELL_HOST, 1134 VIRTIO_BALLOON_F_STATS_VQ, 1135 VIRTIO_BALLOON_F_DEFLATE_ON_OOM, 1136 VIRTIO_BALLOON_F_FREE_PAGE_HINT, 1137 VIRTIO_BALLOON_F_PAGE_POISON, 1138 VIRTIO_BALLOON_F_REPORTING, 1139}; 1140 1141static struct virtio_driver virtio_balloon_driver = { 1142 .feature_table = features, 1143 .feature_table_size = ARRAY_SIZE(features), 1144 .driver.name = KBUILD_MODNAME, 1145 .driver.owner = THIS_MODULE, 1146 .id_table = id_table, 1147 .validate = virtballoon_validate, 1148 .probe = virtballoon_probe, 1149 .remove = virtballoon_remove, 1150 .config_changed = virtballoon_changed, 1151#ifdef CONFIG_PM_SLEEP 1152 .freeze = virtballoon_freeze, 1153 .restore = virtballoon_restore, 1154#endif 1155}; 1156 1157module_virtio_driver(virtio_balloon_driver); 1158MODULE_DEVICE_TABLE(virtio, id_table); 1159MODULE_DESCRIPTION("Virtio balloon driver"); 1160MODULE_LICENSE("GPL"); 1161