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