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/page_reporting.h> 21 22 /* 23 * Balloon device works in 4K page units. So each page is pointed to by 24 * multiple balloon pages. All memory counters in this driver are in balloon 25 * page units. 26 */ 27 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned int)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT) 28 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256 29 /* Maximum number of (4k) pages to deflate on OOM notifications. */ 30 #define VIRTIO_BALLOON_OOM_NR_PAGES 256 31 #define VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY 80 32 33 #define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \ 34 __GFP_NOMEMALLOC) 35 /* The order of free page blocks to report to host */ 36 #define VIRTIO_BALLOON_HINT_BLOCK_ORDER MAX_PAGE_ORDER 37 /* The size of a free page block in bytes */ 38 #define VIRTIO_BALLOON_HINT_BLOCK_BYTES \ 39 (1 << (VIRTIO_BALLOON_HINT_BLOCK_ORDER + PAGE_SHIFT)) 40 #define VIRTIO_BALLOON_HINT_BLOCK_PAGES (1 << VIRTIO_BALLOON_HINT_BLOCK_ORDER) 41 42 enum virtio_balloon_vq { 43 VIRTIO_BALLOON_VQ_INFLATE, 44 VIRTIO_BALLOON_VQ_DEFLATE, 45 VIRTIO_BALLOON_VQ_STATS, 46 VIRTIO_BALLOON_VQ_FREE_PAGE, 47 VIRTIO_BALLOON_VQ_REPORTING, 48 VIRTIO_BALLOON_VQ_MAX 49 }; 50 51 enum virtio_balloon_config_read { 52 VIRTIO_BALLOON_CONFIG_READ_CMD_ID = 0, 53 }; 54 55 struct virtio_balloon { 56 struct virtio_device *vdev; 57 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq; 58 59 /* Balloon's own wq for cpu-intensive work items */ 60 struct workqueue_struct *balloon_wq; 61 /* The free page reporting work item submitted to the balloon wq */ 62 struct work_struct report_free_page_work; 63 64 /* The balloon servicing is delegated to a freezable workqueue. */ 65 struct work_struct update_balloon_stats_work; 66 struct work_struct update_balloon_size_work; 67 68 /* Prevent updating balloon when it is being canceled. */ 69 spinlock_t stop_update_lock; 70 bool stop_update; 71 /* Bitmap to indicate if reading the related config fields are needed */ 72 unsigned long config_read_bitmap; 73 74 /* The list of allocated free pages, waiting to be given back to mm */ 75 struct list_head free_page_list; 76 spinlock_t free_page_list_lock; 77 /* The number of free page blocks on the above list */ 78 unsigned long num_free_page_blocks; 79 /* 80 * The cmd id received from host. 81 * Read it via virtio_balloon_cmd_id_received to get the latest value 82 * sent from host. 83 */ 84 u32 cmd_id_received_cache; 85 /* The cmd id that is actively in use */ 86 __virtio32 cmd_id_active; 87 /* Buffer to store the stop sign */ 88 __virtio32 cmd_id_stop; 89 90 /* Waiting for host to ack the pages we released. */ 91 wait_queue_head_t acked; 92 93 /* Number of balloon pages we've told the Host we're not using. */ 94 unsigned int num_pages; 95 /* 96 * The pages we've told the Host we're not using are enqueued 97 * at vb_dev_info->pages list. 98 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE 99 * to num_pages above. 100 */ 101 struct balloon_dev_info vb_dev_info; 102 103 /* Synchronize access/update to this struct virtio_balloon elements */ 104 struct mutex balloon_lock; 105 106 /* The array of pfns we tell the Host about. */ 107 unsigned int num_pfns; 108 __virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX]; 109 110 /* Memory statistics */ 111 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR]; 112 113 /* Shrinker to return free pages - VIRTIO_BALLOON_F_FREE_PAGE_HINT */ 114 struct shrinker *shrinker; 115 116 /* OOM notifier to deflate on OOM - VIRTIO_BALLOON_F_DEFLATE_ON_OOM */ 117 struct notifier_block oom_nb; 118 119 /* Free page reporting device */ 120 struct virtqueue *reporting_vq; 121 struct page_reporting_dev_info pr_dev_info; 122 123 /* State for keeping the wakeup_source active while adjusting the balloon */ 124 spinlock_t wakeup_lock; 125 bool processing_wakeup_event; 126 u32 wakeup_signal_mask; 127 }; 128 129 #define VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST (1 << 0) 130 #define VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS (1 << 1) 131 132 static const struct virtio_device_id id_table[] = { 133 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID }, 134 { 0 }, 135 }; 136 137 static u32 page_to_balloon_pfn(struct page *page) 138 { 139 unsigned long pfn = page_to_pfn(page); 140 141 BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT); 142 /* Convert pfn from Linux page size to balloon page size. */ 143 return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE; 144 } 145 146 static void start_wakeup_event(struct virtio_balloon *vb, u32 mask) 147 { 148 unsigned long flags; 149 150 spin_lock_irqsave(&vb->wakeup_lock, flags); 151 vb->wakeup_signal_mask |= mask; 152 if (!vb->processing_wakeup_event) { 153 vb->processing_wakeup_event = true; 154 pm_stay_awake(&vb->vdev->dev); 155 } 156 spin_unlock_irqrestore(&vb->wakeup_lock, flags); 157 } 158 159 static void process_wakeup_event(struct virtio_balloon *vb, u32 mask) 160 { 161 spin_lock_irq(&vb->wakeup_lock); 162 vb->wakeup_signal_mask &= ~mask; 163 spin_unlock_irq(&vb->wakeup_lock); 164 } 165 166 static void finish_wakeup_event(struct virtio_balloon *vb) 167 { 168 spin_lock_irq(&vb->wakeup_lock); 169 if (!vb->wakeup_signal_mask && vb->processing_wakeup_event) { 170 vb->processing_wakeup_event = false; 171 pm_relax(&vb->vdev->dev); 172 } 173 spin_unlock_irq(&vb->wakeup_lock); 174 } 175 176 static void balloon_ack(struct virtqueue *vq) 177 { 178 struct virtio_balloon *vb = vq->vdev->priv; 179 180 wake_up(&vb->acked); 181 } 182 183 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq) 184 { 185 struct scatterlist sg; 186 unsigned int len; 187 188 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns); 189 190 /* We should always be able to add one buffer to an empty queue. */ 191 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); 192 virtqueue_kick(vq); 193 194 /* When host has read buffer, this completes via balloon_ack */ 195 wait_event(vb->acked, virtqueue_get_buf(vq, &len)); 196 197 } 198 199 static int virtballoon_free_page_report(struct page_reporting_dev_info *pr_dev_info, 200 struct scatterlist *sg, unsigned int nents) 201 { 202 struct virtio_balloon *vb = 203 container_of(pr_dev_info, struct virtio_balloon, pr_dev_info); 204 struct virtqueue *vq = vb->reporting_vq; 205 unsigned int unused, err; 206 207 /* We should always be able to add these buffers to an empty queue. */ 208 err = virtqueue_add_inbuf(vq, sg, nents, vb, GFP_NOWAIT | __GFP_NOWARN); 209 210 /* 211 * In the extremely unlikely case that something has occurred and we 212 * are able to trigger an error we will simply display a warning 213 * and exit without actually processing the pages. 214 */ 215 if (WARN_ON_ONCE(err)) 216 return err; 217 218 virtqueue_kick(vq); 219 220 /* When host has read buffer, this completes via balloon_ack */ 221 wait_event(vb->acked, virtqueue_get_buf(vq, &unused)); 222 223 return 0; 224 } 225 226 static void set_page_pfns(struct virtio_balloon *vb, 227 __virtio32 pfns[], struct page *page) 228 { 229 unsigned int i; 230 231 BUILD_BUG_ON(VIRTIO_BALLOON_PAGES_PER_PAGE > VIRTIO_BALLOON_ARRAY_PFNS_MAX); 232 233 /* 234 * Set balloon pfns pointing at this page. 235 * Note that the first pfn points at start of the page. 236 */ 237 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++) 238 pfns[i] = cpu_to_virtio32(vb->vdev, 239 page_to_balloon_pfn(page) + i); 240 } 241 242 static unsigned int fill_balloon(struct virtio_balloon *vb, size_t num) 243 { 244 unsigned int num_allocated_pages; 245 unsigned int num_pfns; 246 struct page *page; 247 LIST_HEAD(pages); 248 249 /* We can only do one array worth at a time. */ 250 num = min(num, ARRAY_SIZE(vb->pfns)); 251 252 for (num_pfns = 0; num_pfns < num; 253 num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { 254 struct page *page = balloon_page_alloc(); 255 256 if (!page) { 257 dev_info_ratelimited(&vb->vdev->dev, 258 "Out of puff! Can't get %u pages\n", 259 VIRTIO_BALLOON_PAGES_PER_PAGE); 260 /* Sleep for at least 1/5 of a second before retry. */ 261 msleep(200); 262 break; 263 } 264 265 balloon_page_push(&pages, page); 266 } 267 268 mutex_lock(&vb->balloon_lock); 269 270 vb->num_pfns = 0; 271 272 while ((page = balloon_page_pop(&pages))) { 273 balloon_page_enqueue(&vb->vb_dev_info, page); 274 275 set_page_pfns(vb, vb->pfns + vb->num_pfns, page); 276 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE; 277 if (!virtio_has_feature(vb->vdev, 278 VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 279 adjust_managed_page_count(page, -1); 280 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE; 281 } 282 283 num_allocated_pages = vb->num_pfns; 284 /* Did we get any? */ 285 if (vb->num_pfns != 0) 286 tell_host(vb, vb->inflate_vq); 287 mutex_unlock(&vb->balloon_lock); 288 289 return num_allocated_pages; 290 } 291 292 static void release_pages_balloon(struct virtio_balloon *vb, 293 struct list_head *pages) 294 { 295 struct page *page, *next; 296 297 list_for_each_entry_safe(page, next, pages, lru) { 298 if (!virtio_has_feature(vb->vdev, 299 VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 300 adjust_managed_page_count(page, 1); 301 list_del(&page->lru); 302 put_page(page); /* balloon reference */ 303 } 304 } 305 306 static unsigned int leak_balloon(struct virtio_balloon *vb, size_t num) 307 { 308 unsigned int num_freed_pages; 309 struct page *page; 310 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info; 311 LIST_HEAD(pages); 312 313 /* We can only do one array worth at a time. */ 314 num = min(num, ARRAY_SIZE(vb->pfns)); 315 316 mutex_lock(&vb->balloon_lock); 317 /* We can't release more pages than taken */ 318 num = min(num, (size_t)vb->num_pages); 319 for (vb->num_pfns = 0; vb->num_pfns < num; 320 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) { 321 page = balloon_page_dequeue(vb_dev_info); 322 if (!page) 323 break; 324 set_page_pfns(vb, vb->pfns + vb->num_pfns, page); 325 list_add(&page->lru, &pages); 326 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE; 327 } 328 329 num_freed_pages = vb->num_pfns; 330 /* 331 * Note that if 332 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST); 333 * is true, we *have* to do it in this order 334 */ 335 if (vb->num_pfns != 0) 336 tell_host(vb, vb->deflate_vq); 337 release_pages_balloon(vb, &pages); 338 mutex_unlock(&vb->balloon_lock); 339 return num_freed_pages; 340 } 341 342 static inline void update_stat(struct virtio_balloon *vb, int idx, 343 u16 tag, u64 val) 344 { 345 BUG_ON(idx >= VIRTIO_BALLOON_S_NR); 346 vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag); 347 vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val); 348 } 349 350 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT) 351 352 static unsigned int update_balloon_stats(struct virtio_balloon *vb) 353 { 354 unsigned long events[NR_VM_EVENT_ITEMS]; 355 struct sysinfo i; 356 unsigned int idx = 0; 357 long available; 358 unsigned long caches; 359 360 all_vm_events(events); 361 si_meminfo(&i); 362 363 available = si_mem_available(); 364 caches = global_node_page_state(NR_FILE_PAGES); 365 366 #ifdef CONFIG_VM_EVENT_COUNTERS 367 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN, 368 pages_to_bytes(events[PSWPIN])); 369 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT, 370 pages_to_bytes(events[PSWPOUT])); 371 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]); 372 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]); 373 #ifdef CONFIG_HUGETLB_PAGE 374 update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC, 375 events[HTLB_BUDDY_PGALLOC]); 376 update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL, 377 events[HTLB_BUDDY_PGALLOC_FAIL]); 378 #endif 379 #endif 380 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE, 381 pages_to_bytes(i.freeram)); 382 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT, 383 pages_to_bytes(i.totalram)); 384 update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL, 385 pages_to_bytes(available)); 386 update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES, 387 pages_to_bytes(caches)); 388 389 return idx; 390 } 391 392 /* 393 * While most virtqueues communicate guest-initiated requests to the hypervisor, 394 * the stats queue operates in reverse. The driver initializes the virtqueue 395 * with a single buffer. From that point forward, all conversations consist of 396 * a hypervisor request (a call to this function) which directs us to refill 397 * the virtqueue with a fresh stats buffer. Since stats collection can sleep, 398 * we delegate the job to a freezable workqueue that will do the actual work via 399 * stats_handle_request(). 400 */ 401 static void stats_request(struct virtqueue *vq) 402 { 403 struct virtio_balloon *vb = vq->vdev->priv; 404 405 spin_lock(&vb->stop_update_lock); 406 if (!vb->stop_update) { 407 start_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS); 408 queue_work(system_freezable_wq, &vb->update_balloon_stats_work); 409 } 410 spin_unlock(&vb->stop_update_lock); 411 } 412 413 static void stats_handle_request(struct virtio_balloon *vb) 414 { 415 struct virtqueue *vq; 416 struct scatterlist sg; 417 unsigned int len, num_stats; 418 419 num_stats = update_balloon_stats(vb); 420 421 vq = vb->stats_vq; 422 if (!virtqueue_get_buf(vq, &len)) 423 return; 424 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats); 425 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL); 426 virtqueue_kick(vq); 427 } 428 429 static inline s64 towards_target(struct virtio_balloon *vb) 430 { 431 s64 target; 432 u32 num_pages; 433 434 /* Legacy balloon config space is LE, unlike all other devices. */ 435 virtio_cread_le(vb->vdev, struct virtio_balloon_config, num_pages, 436 &num_pages); 437 438 /* 439 * Aligned up to guest page size to avoid inflating and deflating 440 * balloon endlessly. 441 */ 442 target = ALIGN(num_pages, VIRTIO_BALLOON_PAGES_PER_PAGE); 443 return target - vb->num_pages; 444 } 445 446 /* Gives back @num_to_return blocks of free pages to mm. */ 447 static unsigned long return_free_pages_to_mm(struct virtio_balloon *vb, 448 unsigned long num_to_return) 449 { 450 struct page *page; 451 unsigned long num_returned; 452 453 spin_lock_irq(&vb->free_page_list_lock); 454 for (num_returned = 0; num_returned < num_to_return; num_returned++) { 455 page = balloon_page_pop(&vb->free_page_list); 456 if (!page) 457 break; 458 free_pages((unsigned long)page_address(page), 459 VIRTIO_BALLOON_HINT_BLOCK_ORDER); 460 } 461 vb->num_free_page_blocks -= num_returned; 462 spin_unlock_irq(&vb->free_page_list_lock); 463 464 return num_returned; 465 } 466 467 static void virtio_balloon_queue_free_page_work(struct virtio_balloon *vb) 468 { 469 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 470 return; 471 472 /* No need to queue the work if the bit was already set. */ 473 if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID, 474 &vb->config_read_bitmap)) 475 return; 476 477 queue_work(vb->balloon_wq, &vb->report_free_page_work); 478 } 479 480 static void start_update_balloon_size(struct virtio_balloon *vb) 481 { 482 start_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST); 483 queue_work(system_freezable_wq, &vb->update_balloon_size_work); 484 } 485 486 static void virtballoon_changed(struct virtio_device *vdev) 487 { 488 struct virtio_balloon *vb = vdev->priv; 489 unsigned long flags; 490 491 spin_lock_irqsave(&vb->stop_update_lock, flags); 492 if (!vb->stop_update) { 493 start_update_balloon_size(vb); 494 virtio_balloon_queue_free_page_work(vb); 495 } 496 spin_unlock_irqrestore(&vb->stop_update_lock, flags); 497 } 498 499 static void update_balloon_size(struct virtio_balloon *vb) 500 { 501 u32 actual = vb->num_pages; 502 503 /* Legacy balloon config space is LE, unlike all other devices. */ 504 virtio_cwrite_le(vb->vdev, struct virtio_balloon_config, actual, 505 &actual); 506 } 507 508 static void update_balloon_stats_func(struct work_struct *work) 509 { 510 struct virtio_balloon *vb; 511 512 vb = container_of(work, struct virtio_balloon, 513 update_balloon_stats_work); 514 515 process_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_STATS); 516 stats_handle_request(vb); 517 finish_wakeup_event(vb); 518 } 519 520 static void update_balloon_size_func(struct work_struct *work) 521 { 522 struct virtio_balloon *vb; 523 s64 diff; 524 525 vb = container_of(work, struct virtio_balloon, 526 update_balloon_size_work); 527 528 process_wakeup_event(vb, VIRTIO_BALLOON_WAKEUP_SIGNAL_ADJUST); 529 530 diff = towards_target(vb); 531 532 if (diff) { 533 if (diff > 0) 534 diff -= fill_balloon(vb, diff); 535 else 536 diff += leak_balloon(vb, -diff); 537 update_balloon_size(vb); 538 } 539 540 if (diff) 541 queue_work(system_freezable_wq, work); 542 else 543 finish_wakeup_event(vb); 544 } 545 546 static int init_vqs(struct virtio_balloon *vb) 547 { 548 struct virtqueue *vqs[VIRTIO_BALLOON_VQ_MAX]; 549 vq_callback_t *callbacks[VIRTIO_BALLOON_VQ_MAX]; 550 const char *names[VIRTIO_BALLOON_VQ_MAX]; 551 int err; 552 553 /* 554 * Inflateq and deflateq are used unconditionally. The names[] 555 * will be NULL if the related feature is not enabled, which will 556 * cause no allocation for the corresponding virtqueue in find_vqs. 557 */ 558 callbacks[VIRTIO_BALLOON_VQ_INFLATE] = balloon_ack; 559 names[VIRTIO_BALLOON_VQ_INFLATE] = "inflate"; 560 callbacks[VIRTIO_BALLOON_VQ_DEFLATE] = balloon_ack; 561 names[VIRTIO_BALLOON_VQ_DEFLATE] = "deflate"; 562 callbacks[VIRTIO_BALLOON_VQ_STATS] = NULL; 563 names[VIRTIO_BALLOON_VQ_STATS] = NULL; 564 callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL; 565 names[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL; 566 names[VIRTIO_BALLOON_VQ_REPORTING] = NULL; 567 568 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { 569 names[VIRTIO_BALLOON_VQ_STATS] = "stats"; 570 callbacks[VIRTIO_BALLOON_VQ_STATS] = stats_request; 571 } 572 573 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { 574 names[VIRTIO_BALLOON_VQ_FREE_PAGE] = "free_page_vq"; 575 callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL; 576 } 577 578 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) { 579 names[VIRTIO_BALLOON_VQ_REPORTING] = "reporting_vq"; 580 callbacks[VIRTIO_BALLOON_VQ_REPORTING] = balloon_ack; 581 } 582 583 err = virtio_find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX, vqs, 584 callbacks, names, NULL); 585 if (err) 586 return err; 587 588 vb->inflate_vq = vqs[VIRTIO_BALLOON_VQ_INFLATE]; 589 vb->deflate_vq = vqs[VIRTIO_BALLOON_VQ_DEFLATE]; 590 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) { 591 struct scatterlist sg; 592 unsigned int num_stats; 593 vb->stats_vq = vqs[VIRTIO_BALLOON_VQ_STATS]; 594 595 /* 596 * Prime this virtqueue with one buffer so the hypervisor can 597 * use it to signal us later (it can't be broken yet!). 598 */ 599 num_stats = update_balloon_stats(vb); 600 601 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats); 602 err = virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, 603 GFP_KERNEL); 604 if (err) { 605 dev_warn(&vb->vdev->dev, "%s: add stat_vq failed\n", 606 __func__); 607 return err; 608 } 609 virtqueue_kick(vb->stats_vq); 610 } 611 612 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 613 vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE]; 614 615 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) 616 vb->reporting_vq = vqs[VIRTIO_BALLOON_VQ_REPORTING]; 617 618 return 0; 619 } 620 621 static u32 virtio_balloon_cmd_id_received(struct virtio_balloon *vb) 622 { 623 if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID, 624 &vb->config_read_bitmap)) { 625 /* Legacy balloon config space is LE, unlike all other devices. */ 626 virtio_cread_le(vb->vdev, struct virtio_balloon_config, 627 free_page_hint_cmd_id, 628 &vb->cmd_id_received_cache); 629 } 630 631 return vb->cmd_id_received_cache; 632 } 633 634 static int send_cmd_id_start(struct virtio_balloon *vb) 635 { 636 struct scatterlist sg; 637 struct virtqueue *vq = vb->free_page_vq; 638 int err, unused; 639 640 /* Detach all the used buffers from the vq */ 641 while (virtqueue_get_buf(vq, &unused)) 642 ; 643 644 vb->cmd_id_active = cpu_to_virtio32(vb->vdev, 645 virtio_balloon_cmd_id_received(vb)); 646 sg_init_one(&sg, &vb->cmd_id_active, sizeof(vb->cmd_id_active)); 647 err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_active, GFP_KERNEL); 648 if (!err) 649 virtqueue_kick(vq); 650 return err; 651 } 652 653 static int send_cmd_id_stop(struct virtio_balloon *vb) 654 { 655 struct scatterlist sg; 656 struct virtqueue *vq = vb->free_page_vq; 657 int err, unused; 658 659 /* Detach all the used buffers from the vq */ 660 while (virtqueue_get_buf(vq, &unused)) 661 ; 662 663 sg_init_one(&sg, &vb->cmd_id_stop, sizeof(vb->cmd_id_stop)); 664 err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_stop, GFP_KERNEL); 665 if (!err) 666 virtqueue_kick(vq); 667 return err; 668 } 669 670 static int get_free_page_and_send(struct virtio_balloon *vb) 671 { 672 struct virtqueue *vq = vb->free_page_vq; 673 struct page *page; 674 struct scatterlist sg; 675 int err, unused; 676 void *p; 677 678 /* Detach all the used buffers from the vq */ 679 while (virtqueue_get_buf(vq, &unused)) 680 ; 681 682 page = alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG, 683 VIRTIO_BALLOON_HINT_BLOCK_ORDER); 684 /* 685 * When the allocation returns NULL, it indicates that we have got all 686 * the possible free pages, so return -EINTR to stop. 687 */ 688 if (!page) 689 return -EINTR; 690 691 p = page_address(page); 692 sg_init_one(&sg, p, VIRTIO_BALLOON_HINT_BLOCK_BYTES); 693 /* There is always 1 entry reserved for the cmd id to use. */ 694 if (vq->num_free > 1) { 695 err = virtqueue_add_inbuf(vq, &sg, 1, p, GFP_KERNEL); 696 if (unlikely(err)) { 697 free_pages((unsigned long)p, 698 VIRTIO_BALLOON_HINT_BLOCK_ORDER); 699 return err; 700 } 701 virtqueue_kick(vq); 702 spin_lock_irq(&vb->free_page_list_lock); 703 balloon_page_push(&vb->free_page_list, page); 704 vb->num_free_page_blocks++; 705 spin_unlock_irq(&vb->free_page_list_lock); 706 } else { 707 /* 708 * The vq has no available entry to add this page block, so 709 * just free it. 710 */ 711 free_pages((unsigned long)p, VIRTIO_BALLOON_HINT_BLOCK_ORDER); 712 } 713 714 return 0; 715 } 716 717 static int send_free_pages(struct virtio_balloon *vb) 718 { 719 int err; 720 u32 cmd_id_active; 721 722 while (1) { 723 /* 724 * If a stop id or a new cmd id was just received from host, 725 * stop the reporting. 726 */ 727 cmd_id_active = virtio32_to_cpu(vb->vdev, vb->cmd_id_active); 728 if (unlikely(cmd_id_active != 729 virtio_balloon_cmd_id_received(vb))) 730 break; 731 732 /* 733 * The free page blocks are allocated and sent to host one by 734 * one. 735 */ 736 err = get_free_page_and_send(vb); 737 if (err == -EINTR) 738 break; 739 else if (unlikely(err)) 740 return err; 741 } 742 743 return 0; 744 } 745 746 static void virtio_balloon_report_free_page(struct virtio_balloon *vb) 747 { 748 int err; 749 struct device *dev = &vb->vdev->dev; 750 751 /* Start by sending the received cmd id to host with an outbuf. */ 752 err = send_cmd_id_start(vb); 753 if (unlikely(err)) 754 dev_err(dev, "Failed to send a start id, err = %d\n", err); 755 756 err = send_free_pages(vb); 757 if (unlikely(err)) 758 dev_err(dev, "Failed to send a free page, err = %d\n", err); 759 760 /* End by sending a stop id to host with an outbuf. */ 761 err = send_cmd_id_stop(vb); 762 if (unlikely(err)) 763 dev_err(dev, "Failed to send a stop id, err = %d\n", err); 764 } 765 766 static void report_free_page_func(struct work_struct *work) 767 { 768 struct virtio_balloon *vb = container_of(work, struct virtio_balloon, 769 report_free_page_work); 770 u32 cmd_id_received; 771 772 cmd_id_received = virtio_balloon_cmd_id_received(vb); 773 if (cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) { 774 /* Pass ULONG_MAX to give back all the free pages */ 775 return_free_pages_to_mm(vb, ULONG_MAX); 776 } else if (cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP && 777 cmd_id_received != 778 virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) { 779 virtio_balloon_report_free_page(vb); 780 } 781 } 782 783 #ifdef CONFIG_BALLOON_COMPACTION 784 /* 785 * virtballoon_migratepage - perform the balloon page migration on behalf of 786 * a compaction thread. (called under page lock) 787 * @vb_dev_info: the balloon device 788 * @newpage: page that will replace the isolated page after migration finishes. 789 * @page : the isolated (old) page that is about to be migrated to newpage. 790 * @mode : compaction mode -- not used for balloon page migration. 791 * 792 * After a ballooned page gets isolated by compaction procedures, this is the 793 * function that performs the page migration on behalf of a compaction thread 794 * The page migration for virtio balloon is done in a simple swap fashion which 795 * follows these two macro steps: 796 * 1) insert newpage into vb->pages list and update the host about it; 797 * 2) update the host about the old page removed from vb->pages list; 798 * 799 * This function preforms the balloon page migration task. 800 * Called through movable_operations->migrate_page 801 */ 802 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info, 803 struct page *newpage, struct page *page, enum migrate_mode mode) 804 { 805 struct virtio_balloon *vb = container_of(vb_dev_info, 806 struct virtio_balloon, vb_dev_info); 807 unsigned long flags; 808 809 /* 810 * In order to avoid lock contention while migrating pages concurrently 811 * to leak_balloon() or fill_balloon() we just give up the balloon_lock 812 * this turn, as it is easier to retry the page migration later. 813 * This also prevents fill_balloon() getting stuck into a mutex 814 * recursion in the case it ends up triggering memory compaction 815 * while it is attempting to inflate the ballon. 816 */ 817 if (!mutex_trylock(&vb->balloon_lock)) 818 return -EAGAIN; 819 820 get_page(newpage); /* balloon reference */ 821 822 /* 823 * When we migrate a page to a different zone and adjusted the 824 * managed page count when inflating, we have to fixup the count of 825 * both involved zones. 826 */ 827 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM) && 828 page_zone(page) != page_zone(newpage)) { 829 adjust_managed_page_count(page, 1); 830 adjust_managed_page_count(newpage, -1); 831 } 832 833 /* balloon's page migration 1st step -- inflate "newpage" */ 834 spin_lock_irqsave(&vb_dev_info->pages_lock, flags); 835 balloon_page_insert(vb_dev_info, newpage); 836 vb_dev_info->isolated_pages--; 837 __count_vm_event(BALLOON_MIGRATE); 838 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags); 839 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; 840 set_page_pfns(vb, vb->pfns, newpage); 841 tell_host(vb, vb->inflate_vq); 842 843 /* balloon's page migration 2nd step -- deflate "page" */ 844 spin_lock_irqsave(&vb_dev_info->pages_lock, flags); 845 balloon_page_delete(page); 846 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags); 847 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE; 848 set_page_pfns(vb, vb->pfns, page); 849 tell_host(vb, vb->deflate_vq); 850 851 mutex_unlock(&vb->balloon_lock); 852 853 put_page(page); /* balloon reference */ 854 855 return MIGRATEPAGE_SUCCESS; 856 } 857 #endif /* CONFIG_BALLOON_COMPACTION */ 858 859 static unsigned long shrink_free_pages(struct virtio_balloon *vb, 860 unsigned long pages_to_free) 861 { 862 unsigned long blocks_to_free, blocks_freed; 863 864 pages_to_free = round_up(pages_to_free, 865 VIRTIO_BALLOON_HINT_BLOCK_PAGES); 866 blocks_to_free = pages_to_free / VIRTIO_BALLOON_HINT_BLOCK_PAGES; 867 blocks_freed = return_free_pages_to_mm(vb, blocks_to_free); 868 869 return blocks_freed * VIRTIO_BALLOON_HINT_BLOCK_PAGES; 870 } 871 872 static unsigned long virtio_balloon_shrinker_scan(struct shrinker *shrinker, 873 struct shrink_control *sc) 874 { 875 struct virtio_balloon *vb = shrinker->private_data; 876 877 return shrink_free_pages(vb, sc->nr_to_scan); 878 } 879 880 static unsigned long virtio_balloon_shrinker_count(struct shrinker *shrinker, 881 struct shrink_control *sc) 882 { 883 struct virtio_balloon *vb = shrinker->private_data; 884 885 return vb->num_free_page_blocks * VIRTIO_BALLOON_HINT_BLOCK_PAGES; 886 } 887 888 static int virtio_balloon_oom_notify(struct notifier_block *nb, 889 unsigned long dummy, void *parm) 890 { 891 struct virtio_balloon *vb = container_of(nb, 892 struct virtio_balloon, oom_nb); 893 unsigned long *freed = parm; 894 895 *freed += leak_balloon(vb, VIRTIO_BALLOON_OOM_NR_PAGES) / 896 VIRTIO_BALLOON_PAGES_PER_PAGE; 897 update_balloon_size(vb); 898 899 return NOTIFY_OK; 900 } 901 902 static void virtio_balloon_unregister_shrinker(struct virtio_balloon *vb) 903 { 904 shrinker_free(vb->shrinker); 905 } 906 907 static int virtio_balloon_register_shrinker(struct virtio_balloon *vb) 908 { 909 vb->shrinker = shrinker_alloc(0, "virtio-balloon"); 910 if (!vb->shrinker) 911 return -ENOMEM; 912 913 vb->shrinker->scan_objects = virtio_balloon_shrinker_scan; 914 vb->shrinker->count_objects = virtio_balloon_shrinker_count; 915 vb->shrinker->private_data = vb; 916 917 shrinker_register(vb->shrinker); 918 919 return 0; 920 } 921 922 static int virtballoon_probe(struct virtio_device *vdev) 923 { 924 struct virtio_balloon *vb; 925 int err; 926 927 if (!vdev->config->get) { 928 dev_err(&vdev->dev, "%s failure: config access disabled\n", 929 __func__); 930 return -EINVAL; 931 } 932 933 vdev->priv = vb = kzalloc(sizeof(*vb), GFP_KERNEL); 934 if (!vb) { 935 err = -ENOMEM; 936 goto out; 937 } 938 939 INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func); 940 INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func); 941 spin_lock_init(&vb->stop_update_lock); 942 mutex_init(&vb->balloon_lock); 943 init_waitqueue_head(&vb->acked); 944 vb->vdev = vdev; 945 946 balloon_devinfo_init(&vb->vb_dev_info); 947 948 err = init_vqs(vb); 949 if (err) 950 goto out_free_vb; 951 952 #ifdef CONFIG_BALLOON_COMPACTION 953 vb->vb_dev_info.migratepage = virtballoon_migratepage; 954 #endif 955 if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { 956 /* 957 * There is always one entry reserved for cmd id, so the ring 958 * size needs to be at least two to report free page hints. 959 */ 960 if (virtqueue_get_vring_size(vb->free_page_vq) < 2) { 961 err = -ENOSPC; 962 goto out_del_vqs; 963 } 964 vb->balloon_wq = alloc_workqueue("balloon-wq", 965 WQ_FREEZABLE | WQ_CPU_INTENSIVE, 0); 966 if (!vb->balloon_wq) { 967 err = -ENOMEM; 968 goto out_del_vqs; 969 } 970 INIT_WORK(&vb->report_free_page_work, report_free_page_func); 971 vb->cmd_id_received_cache = VIRTIO_BALLOON_CMD_ID_STOP; 972 vb->cmd_id_active = cpu_to_virtio32(vb->vdev, 973 VIRTIO_BALLOON_CMD_ID_STOP); 974 vb->cmd_id_stop = cpu_to_virtio32(vb->vdev, 975 VIRTIO_BALLOON_CMD_ID_STOP); 976 spin_lock_init(&vb->free_page_list_lock); 977 INIT_LIST_HEAD(&vb->free_page_list); 978 /* 979 * We're allowed to reuse any free pages, even if they are 980 * still to be processed by the host. 981 */ 982 err = virtio_balloon_register_shrinker(vb); 983 if (err) 984 goto out_del_balloon_wq; 985 } 986 987 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) { 988 vb->oom_nb.notifier_call = virtio_balloon_oom_notify; 989 vb->oom_nb.priority = VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY; 990 err = register_oom_notifier(&vb->oom_nb); 991 if (err < 0) 992 goto out_unregister_shrinker; 993 } 994 995 if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) { 996 /* Start with poison val of 0 representing general init */ 997 __u32 poison_val = 0; 998 999 /* 1000 * Let the hypervisor know that we are expecting a 1001 * specific value to be written back in balloon pages. 1002 * 1003 * If the PAGE_POISON value was larger than a byte we would 1004 * need to byte swap poison_val here to guarantee it is 1005 * little-endian. However for now it is a single byte so we 1006 * can pass it as-is. 1007 */ 1008 if (!want_init_on_free()) 1009 memset(&poison_val, PAGE_POISON, sizeof(poison_val)); 1010 1011 virtio_cwrite_le(vb->vdev, struct virtio_balloon_config, 1012 poison_val, &poison_val); 1013 } 1014 1015 vb->pr_dev_info.report = virtballoon_free_page_report; 1016 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) { 1017 unsigned int capacity; 1018 1019 capacity = virtqueue_get_vring_size(vb->reporting_vq); 1020 if (capacity < PAGE_REPORTING_CAPACITY) { 1021 err = -ENOSPC; 1022 goto out_unregister_oom; 1023 } 1024 1025 /* 1026 * The default page reporting order is @pageblock_order, which 1027 * corresponds to 512MB in size on ARM64 when 64KB base page 1028 * size is used. The page reporting won't be triggered if the 1029 * freeing page can't come up with a free area like that huge. 1030 * So we specify the page reporting order to 5, corresponding 1031 * to 2MB. It helps to avoid THP splitting if 4KB base page 1032 * size is used by host. 1033 * 1034 * Ideally, the page reporting order is selected based on the 1035 * host's base page size. However, it needs more work to report 1036 * that value. The hard-coded order would be fine currently. 1037 */ 1038 #if defined(CONFIG_ARM64) && defined(CONFIG_ARM64_64K_PAGES) 1039 vb->pr_dev_info.order = 5; 1040 #endif 1041 1042 err = page_reporting_register(&vb->pr_dev_info); 1043 if (err) 1044 goto out_unregister_oom; 1045 } 1046 1047 spin_lock_init(&vb->wakeup_lock); 1048 1049 /* 1050 * The virtio balloon itself can't wake up the device, but it is 1051 * responsible for processing wakeup events passed up from the transport 1052 * layer. Wakeup sources don't support nesting/chaining calls, so we use 1053 * our own wakeup source to ensure wakeup events are properly handled 1054 * without trampling on the transport layer's wakeup source. 1055 */ 1056 device_set_wakeup_capable(&vb->vdev->dev, true); 1057 1058 virtio_device_ready(vdev); 1059 1060 if (towards_target(vb)) 1061 virtballoon_changed(vdev); 1062 return 0; 1063 1064 out_unregister_oom: 1065 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 1066 unregister_oom_notifier(&vb->oom_nb); 1067 out_unregister_shrinker: 1068 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 1069 virtio_balloon_unregister_shrinker(vb); 1070 out_del_balloon_wq: 1071 if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 1072 destroy_workqueue(vb->balloon_wq); 1073 out_del_vqs: 1074 vdev->config->del_vqs(vdev); 1075 out_free_vb: 1076 kfree(vb); 1077 out: 1078 return err; 1079 } 1080 1081 static void remove_common(struct virtio_balloon *vb) 1082 { 1083 /* There might be pages left in the balloon: free them. */ 1084 while (vb->num_pages) 1085 leak_balloon(vb, vb->num_pages); 1086 update_balloon_size(vb); 1087 1088 /* There might be free pages that are being reported: release them. */ 1089 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 1090 return_free_pages_to_mm(vb, ULONG_MAX); 1091 1092 /* Now we reset the device so we can clean up the queues. */ 1093 virtio_reset_device(vb->vdev); 1094 1095 vb->vdev->config->del_vqs(vb->vdev); 1096 } 1097 1098 static void virtballoon_remove(struct virtio_device *vdev) 1099 { 1100 struct virtio_balloon *vb = vdev->priv; 1101 1102 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) 1103 page_reporting_unregister(&vb->pr_dev_info); 1104 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) 1105 unregister_oom_notifier(&vb->oom_nb); 1106 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) 1107 virtio_balloon_unregister_shrinker(vb); 1108 spin_lock_irq(&vb->stop_update_lock); 1109 vb->stop_update = true; 1110 spin_unlock_irq(&vb->stop_update_lock); 1111 cancel_work_sync(&vb->update_balloon_size_work); 1112 cancel_work_sync(&vb->update_balloon_stats_work); 1113 1114 if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { 1115 cancel_work_sync(&vb->report_free_page_work); 1116 destroy_workqueue(vb->balloon_wq); 1117 } 1118 1119 remove_common(vb); 1120 kfree(vb); 1121 } 1122 1123 #ifdef CONFIG_PM_SLEEP 1124 static int virtballoon_freeze(struct virtio_device *vdev) 1125 { 1126 struct virtio_balloon *vb = vdev->priv; 1127 1128 /* 1129 * The workqueue is already frozen by the PM core before this 1130 * function is called. 1131 */ 1132 remove_common(vb); 1133 return 0; 1134 } 1135 1136 static int virtballoon_restore(struct virtio_device *vdev) 1137 { 1138 struct virtio_balloon *vb = vdev->priv; 1139 int ret; 1140 1141 ret = init_vqs(vdev->priv); 1142 if (ret) 1143 return ret; 1144 1145 virtio_device_ready(vdev); 1146 1147 if (towards_target(vb)) 1148 virtballoon_changed(vdev); 1149 update_balloon_size(vb); 1150 return 0; 1151 } 1152 #endif 1153 1154 static int virtballoon_validate(struct virtio_device *vdev) 1155 { 1156 /* 1157 * Inform the hypervisor that our pages are poisoned or 1158 * initialized. If we cannot do that then we should disable 1159 * page reporting as it could potentially change the contents 1160 * of our free pages. 1161 */ 1162 if (!want_init_on_free() && !page_poisoning_enabled_static()) 1163 __virtio_clear_bit(vdev, VIRTIO_BALLOON_F_PAGE_POISON); 1164 else if (!virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) 1165 __virtio_clear_bit(vdev, VIRTIO_BALLOON_F_REPORTING); 1166 1167 __virtio_clear_bit(vdev, VIRTIO_F_ACCESS_PLATFORM); 1168 return 0; 1169 } 1170 1171 static unsigned int features[] = { 1172 VIRTIO_BALLOON_F_MUST_TELL_HOST, 1173 VIRTIO_BALLOON_F_STATS_VQ, 1174 VIRTIO_BALLOON_F_DEFLATE_ON_OOM, 1175 VIRTIO_BALLOON_F_FREE_PAGE_HINT, 1176 VIRTIO_BALLOON_F_PAGE_POISON, 1177 VIRTIO_BALLOON_F_REPORTING, 1178 }; 1179 1180 static struct virtio_driver virtio_balloon_driver = { 1181 .feature_table = features, 1182 .feature_table_size = ARRAY_SIZE(features), 1183 .driver.name = KBUILD_MODNAME, 1184 .id_table = id_table, 1185 .validate = virtballoon_validate, 1186 .probe = virtballoon_probe, 1187 .remove = virtballoon_remove, 1188 .config_changed = virtballoon_changed, 1189 #ifdef CONFIG_PM_SLEEP 1190 .freeze = virtballoon_freeze, 1191 .restore = virtballoon_restore, 1192 #endif 1193 }; 1194 1195 module_virtio_driver(virtio_balloon_driver); 1196 MODULE_DEVICE_TABLE(virtio, id_table); 1197 MODULE_DESCRIPTION("Virtio balloon driver"); 1198 MODULE_LICENSE("GPL"); 1199