1 /* 2 * Virtio Balloon Device 3 * 4 * Copyright IBM, Corp. 2008 5 * Copyright (C) 2011 Red Hat, Inc. 6 * Copyright (C) 2011 Amit Shah <amit.shah@redhat.com> 7 * 8 * Authors: 9 * Anthony Liguori <aliguori@us.ibm.com> 10 * 11 * This work is licensed under the terms of the GNU GPL, version 2. See 12 * the COPYING file in the top-level directory. 13 * 14 */ 15 16 #include "qemu/osdep.h" 17 #include "qemu/iov.h" 18 #include "qemu/module.h" 19 #include "qemu/timer.h" 20 #include "hw/virtio/virtio.h" 21 #include "hw/mem/pc-dimm.h" 22 #include "sysemu/balloon.h" 23 #include "hw/virtio/virtio-balloon.h" 24 #include "exec/address-spaces.h" 25 #include "qapi/error.h" 26 #include "qapi/qapi-events-misc.h" 27 #include "qapi/visitor.h" 28 #include "trace.h" 29 #include "qemu/error-report.h" 30 #include "migration/misc.h" 31 32 #include "hw/virtio/virtio-bus.h" 33 #include "hw/virtio/virtio-access.h" 34 35 #define BALLOON_PAGE_SIZE (1 << VIRTIO_BALLOON_PFN_SHIFT) 36 37 typedef struct PartiallyBalloonedPage { 38 ram_addr_t base_gpa; 39 unsigned long *bitmap; 40 } PartiallyBalloonedPage; 41 42 static void virtio_balloon_pbp_free(PartiallyBalloonedPage *pbp) 43 { 44 if (!pbp) { 45 return; 46 } 47 g_free(pbp->bitmap); 48 g_free(pbp); 49 } 50 51 static PartiallyBalloonedPage *virtio_balloon_pbp_alloc(ram_addr_t base_gpa, 52 long subpages) 53 { 54 PartiallyBalloonedPage *pbp = g_new0(PartiallyBalloonedPage, 1); 55 56 pbp->base_gpa = base_gpa; 57 pbp->bitmap = bitmap_new(subpages); 58 59 return pbp; 60 } 61 62 static bool virtio_balloon_pbp_matches(PartiallyBalloonedPage *pbp, 63 ram_addr_t base_gpa) 64 { 65 return pbp->base_gpa == base_gpa; 66 } 67 68 static void balloon_inflate_page(VirtIOBalloon *balloon, 69 MemoryRegion *mr, hwaddr mr_offset, 70 PartiallyBalloonedPage **pbp) 71 { 72 void *addr = memory_region_get_ram_ptr(mr) + mr_offset; 73 ram_addr_t rb_offset, rb_aligned_offset, base_gpa; 74 RAMBlock *rb; 75 size_t rb_page_size; 76 int subpages; 77 78 /* XXX is there a better way to get to the RAMBlock than via a 79 * host address? */ 80 rb = qemu_ram_block_from_host(addr, false, &rb_offset); 81 rb_page_size = qemu_ram_pagesize(rb); 82 83 if (rb_page_size == BALLOON_PAGE_SIZE) { 84 /* Easy case */ 85 86 ram_block_discard_range(rb, rb_offset, rb_page_size); 87 /* We ignore errors from ram_block_discard_range(), because it 88 * has already reported them, and failing to discard a balloon 89 * page is not fatal */ 90 return; 91 } 92 93 /* Hard case 94 * 95 * We've put a piece of a larger host page into the balloon - we 96 * need to keep track until we have a whole host page to 97 * discard 98 */ 99 warn_report_once( 100 "Balloon used with backing page size > 4kiB, this may not be reliable"); 101 102 rb_aligned_offset = QEMU_ALIGN_DOWN(rb_offset, rb_page_size); 103 subpages = rb_page_size / BALLOON_PAGE_SIZE; 104 base_gpa = memory_region_get_ram_addr(mr) + mr_offset - 105 (rb_offset - rb_aligned_offset); 106 107 if (*pbp && !virtio_balloon_pbp_matches(*pbp, base_gpa)) { 108 /* We've partially ballooned part of a host page, but now 109 * we're trying to balloon part of a different one. Too hard, 110 * give up on the old partial page */ 111 virtio_balloon_pbp_free(*pbp); 112 *pbp = NULL; 113 } 114 115 if (!*pbp) { 116 *pbp = virtio_balloon_pbp_alloc(base_gpa, subpages); 117 } 118 119 set_bit((rb_offset - rb_aligned_offset) / BALLOON_PAGE_SIZE, 120 (*pbp)->bitmap); 121 122 if (bitmap_full((*pbp)->bitmap, subpages)) { 123 /* We've accumulated a full host page, we can actually discard 124 * it now */ 125 126 ram_block_discard_range(rb, rb_aligned_offset, rb_page_size); 127 /* We ignore errors from ram_block_discard_range(), because it 128 * has already reported them, and failing to discard a balloon 129 * page is not fatal */ 130 virtio_balloon_pbp_free(*pbp); 131 *pbp = NULL; 132 } 133 } 134 135 static void balloon_deflate_page(VirtIOBalloon *balloon, 136 MemoryRegion *mr, hwaddr mr_offset) 137 { 138 void *addr = memory_region_get_ram_ptr(mr) + mr_offset; 139 ram_addr_t rb_offset; 140 RAMBlock *rb; 141 size_t rb_page_size; 142 void *host_addr; 143 int ret; 144 145 /* XXX is there a better way to get to the RAMBlock than via a 146 * host address? */ 147 rb = qemu_ram_block_from_host(addr, false, &rb_offset); 148 rb_page_size = qemu_ram_pagesize(rb); 149 150 host_addr = (void *)((uintptr_t)addr & ~(rb_page_size - 1)); 151 152 /* When a page is deflated, we hint the whole host page it lives 153 * on, since we can't do anything smaller */ 154 ret = qemu_madvise(host_addr, rb_page_size, QEMU_MADV_WILLNEED); 155 if (ret != 0) { 156 warn_report("Couldn't MADV_WILLNEED on balloon deflate: %s", 157 strerror(errno)); 158 /* Otherwise ignore, failing to page hint shouldn't be fatal */ 159 } 160 } 161 162 static const char *balloon_stat_names[] = { 163 [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in", 164 [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out", 165 [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults", 166 [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults", 167 [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory", 168 [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory", 169 [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory", 170 [VIRTIO_BALLOON_S_CACHES] = "stat-disk-caches", 171 [VIRTIO_BALLOON_S_HTLB_PGALLOC] = "stat-htlb-pgalloc", 172 [VIRTIO_BALLOON_S_HTLB_PGFAIL] = "stat-htlb-pgfail", 173 [VIRTIO_BALLOON_S_NR] = NULL 174 }; 175 176 /* 177 * reset_stats - Mark all items in the stats array as unset 178 * 179 * This function needs to be called at device initialization and before 180 * updating to a set of newly-generated stats. This will ensure that no 181 * stale values stick around in case the guest reports a subset of the supported 182 * statistics. 183 */ 184 static inline void reset_stats(VirtIOBalloon *dev) 185 { 186 int i; 187 for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1); 188 } 189 190 static bool balloon_stats_supported(const VirtIOBalloon *s) 191 { 192 VirtIODevice *vdev = VIRTIO_DEVICE(s); 193 return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ); 194 } 195 196 static bool balloon_stats_enabled(const VirtIOBalloon *s) 197 { 198 return s->stats_poll_interval > 0; 199 } 200 201 static void balloon_stats_destroy_timer(VirtIOBalloon *s) 202 { 203 if (balloon_stats_enabled(s)) { 204 timer_del(s->stats_timer); 205 timer_free(s->stats_timer); 206 s->stats_timer = NULL; 207 s->stats_poll_interval = 0; 208 } 209 } 210 211 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs) 212 { 213 timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000); 214 } 215 216 static void balloon_stats_poll_cb(void *opaque) 217 { 218 VirtIOBalloon *s = opaque; 219 VirtIODevice *vdev = VIRTIO_DEVICE(s); 220 221 if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) { 222 /* re-schedule */ 223 balloon_stats_change_timer(s, s->stats_poll_interval); 224 return; 225 } 226 227 virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset); 228 virtio_notify(vdev, s->svq); 229 g_free(s->stats_vq_elem); 230 s->stats_vq_elem = NULL; 231 } 232 233 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name, 234 void *opaque, Error **errp) 235 { 236 Error *err = NULL; 237 VirtIOBalloon *s = opaque; 238 int i; 239 240 visit_start_struct(v, name, NULL, 0, &err); 241 if (err) { 242 goto out; 243 } 244 visit_type_int(v, "last-update", &s->stats_last_update, &err); 245 if (err) { 246 goto out_end; 247 } 248 249 visit_start_struct(v, "stats", NULL, 0, &err); 250 if (err) { 251 goto out_end; 252 } 253 for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) { 254 visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err); 255 if (err) { 256 goto out_nested; 257 } 258 } 259 visit_check_struct(v, &err); 260 out_nested: 261 visit_end_struct(v, NULL); 262 263 if (!err) { 264 visit_check_struct(v, &err); 265 } 266 out_end: 267 visit_end_struct(v, NULL); 268 out: 269 error_propagate(errp, err); 270 } 271 272 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v, 273 const char *name, void *opaque, 274 Error **errp) 275 { 276 VirtIOBalloon *s = opaque; 277 visit_type_int(v, name, &s->stats_poll_interval, errp); 278 } 279 280 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v, 281 const char *name, void *opaque, 282 Error **errp) 283 { 284 VirtIOBalloon *s = opaque; 285 Error *local_err = NULL; 286 int64_t value; 287 288 visit_type_int(v, name, &value, &local_err); 289 if (local_err) { 290 error_propagate(errp, local_err); 291 return; 292 } 293 294 if (value < 0) { 295 error_setg(errp, "timer value must be greater than zero"); 296 return; 297 } 298 299 if (value > UINT32_MAX) { 300 error_setg(errp, "timer value is too big"); 301 return; 302 } 303 304 if (value == s->stats_poll_interval) { 305 return; 306 } 307 308 if (value == 0) { 309 /* timer=0 disables the timer */ 310 balloon_stats_destroy_timer(s); 311 return; 312 } 313 314 if (balloon_stats_enabled(s)) { 315 /* timer interval change */ 316 s->stats_poll_interval = value; 317 balloon_stats_change_timer(s, value); 318 return; 319 } 320 321 /* create a new timer */ 322 g_assert(s->stats_timer == NULL); 323 s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s); 324 s->stats_poll_interval = value; 325 balloon_stats_change_timer(s, 0); 326 } 327 328 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq) 329 { 330 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 331 PartiallyBalloonedPage *pbp = NULL; 332 VirtQueueElement *elem; 333 MemoryRegionSection section; 334 335 for (;;) { 336 size_t offset = 0; 337 uint32_t pfn; 338 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 339 if (!elem) { 340 break; 341 } 342 343 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) { 344 unsigned int p = virtio_ldl_p(vdev, &pfn); 345 hwaddr pa; 346 347 pa = (hwaddr) p << VIRTIO_BALLOON_PFN_SHIFT; 348 offset += 4; 349 350 section = memory_region_find(get_system_memory(), pa, 351 BALLOON_PAGE_SIZE); 352 if (!section.mr) { 353 trace_virtio_balloon_bad_addr(pa); 354 continue; 355 } 356 if (!memory_region_is_ram(section.mr) || 357 memory_region_is_rom(section.mr) || 358 memory_region_is_romd(section.mr)) { 359 trace_virtio_balloon_bad_addr(pa); 360 memory_region_unref(section.mr); 361 continue; 362 } 363 364 trace_virtio_balloon_handle_output(memory_region_name(section.mr), 365 pa); 366 if (!qemu_balloon_is_inhibited()) { 367 if (vq == s->ivq) { 368 balloon_inflate_page(s, section.mr, 369 section.offset_within_region, &pbp); 370 } else if (vq == s->dvq) { 371 balloon_deflate_page(s, section.mr, section.offset_within_region); 372 } else { 373 g_assert_not_reached(); 374 } 375 } 376 memory_region_unref(section.mr); 377 } 378 379 virtqueue_push(vq, elem, offset); 380 virtio_notify(vdev, vq); 381 g_free(elem); 382 } 383 384 virtio_balloon_pbp_free(pbp); 385 } 386 387 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq) 388 { 389 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 390 VirtQueueElement *elem; 391 VirtIOBalloonStat stat; 392 size_t offset = 0; 393 qemu_timeval tv; 394 395 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 396 if (!elem) { 397 goto out; 398 } 399 400 if (s->stats_vq_elem != NULL) { 401 /* This should never happen if the driver follows the spec. */ 402 virtqueue_push(vq, s->stats_vq_elem, 0); 403 virtio_notify(vdev, vq); 404 g_free(s->stats_vq_elem); 405 } 406 407 s->stats_vq_elem = elem; 408 409 /* Initialize the stats to get rid of any stale values. This is only 410 * needed to handle the case where a guest supports fewer stats than it 411 * used to (ie. it has booted into an old kernel). 412 */ 413 reset_stats(s); 414 415 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat)) 416 == sizeof(stat)) { 417 uint16_t tag = virtio_tswap16(vdev, stat.tag); 418 uint64_t val = virtio_tswap64(vdev, stat.val); 419 420 offset += sizeof(stat); 421 if (tag < VIRTIO_BALLOON_S_NR) 422 s->stats[tag] = val; 423 } 424 s->stats_vq_offset = offset; 425 426 if (qemu_gettimeofday(&tv) < 0) { 427 warn_report("%s: failed to get time of day", __func__); 428 goto out; 429 } 430 431 s->stats_last_update = tv.tv_sec; 432 433 out: 434 if (balloon_stats_enabled(s)) { 435 balloon_stats_change_timer(s, s->stats_poll_interval); 436 } 437 } 438 439 static void virtio_balloon_handle_free_page_vq(VirtIODevice *vdev, 440 VirtQueue *vq) 441 { 442 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 443 qemu_bh_schedule(s->free_page_bh); 444 } 445 446 static bool get_free_page_hints(VirtIOBalloon *dev) 447 { 448 VirtQueueElement *elem; 449 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 450 VirtQueue *vq = dev->free_page_vq; 451 bool ret = true; 452 453 while (dev->block_iothread) { 454 qemu_cond_wait(&dev->free_page_cond, &dev->free_page_lock); 455 } 456 457 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 458 if (!elem) { 459 return false; 460 } 461 462 if (elem->out_num) { 463 uint32_t id; 464 size_t size = iov_to_buf(elem->out_sg, elem->out_num, 0, 465 &id, sizeof(id)); 466 467 virtio_tswap32s(vdev, &id); 468 if (unlikely(size != sizeof(id))) { 469 virtio_error(vdev, "received an incorrect cmd id"); 470 ret = false; 471 goto out; 472 } 473 if (id == dev->free_page_report_cmd_id) { 474 dev->free_page_report_status = FREE_PAGE_REPORT_S_START; 475 } else { 476 /* 477 * Stop the optimization only when it has started. This 478 * avoids a stale stop sign for the previous command. 479 */ 480 if (dev->free_page_report_status == FREE_PAGE_REPORT_S_START) { 481 dev->free_page_report_status = FREE_PAGE_REPORT_S_STOP; 482 } 483 } 484 } 485 486 if (elem->in_num) { 487 if (dev->free_page_report_status == FREE_PAGE_REPORT_S_START) { 488 qemu_guest_free_page_hint(elem->in_sg[0].iov_base, 489 elem->in_sg[0].iov_len); 490 } 491 } 492 493 out: 494 virtqueue_push(vq, elem, 1); 495 g_free(elem); 496 return ret; 497 } 498 499 static void virtio_ballloon_get_free_page_hints(void *opaque) 500 { 501 VirtIOBalloon *dev = opaque; 502 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 503 VirtQueue *vq = dev->free_page_vq; 504 bool continue_to_get_hints; 505 506 do { 507 qemu_mutex_lock(&dev->free_page_lock); 508 virtio_queue_set_notification(vq, 0); 509 continue_to_get_hints = get_free_page_hints(dev); 510 qemu_mutex_unlock(&dev->free_page_lock); 511 virtio_notify(vdev, vq); 512 /* 513 * Start to poll the vq once the reporting started. Otherwise, continue 514 * only when there are entries on the vq, which need to be given back. 515 */ 516 } while (continue_to_get_hints || 517 dev->free_page_report_status == FREE_PAGE_REPORT_S_START); 518 virtio_queue_set_notification(vq, 1); 519 } 520 521 static bool virtio_balloon_free_page_support(void *opaque) 522 { 523 VirtIOBalloon *s = opaque; 524 VirtIODevice *vdev = VIRTIO_DEVICE(s); 525 526 return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT); 527 } 528 529 static void virtio_balloon_free_page_start(VirtIOBalloon *s) 530 { 531 VirtIODevice *vdev = VIRTIO_DEVICE(s); 532 533 /* For the stop and copy phase, we don't need to start the optimization */ 534 if (!vdev->vm_running) { 535 return; 536 } 537 538 if (s->free_page_report_cmd_id == UINT_MAX) { 539 s->free_page_report_cmd_id = 540 VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN; 541 } else { 542 s->free_page_report_cmd_id++; 543 } 544 545 s->free_page_report_status = FREE_PAGE_REPORT_S_REQUESTED; 546 virtio_notify_config(vdev); 547 } 548 549 static void virtio_balloon_free_page_stop(VirtIOBalloon *s) 550 { 551 VirtIODevice *vdev = VIRTIO_DEVICE(s); 552 553 if (s->free_page_report_status != FREE_PAGE_REPORT_S_STOP) { 554 /* 555 * The lock also guarantees us that the 556 * virtio_ballloon_get_free_page_hints exits after the 557 * free_page_report_status is set to S_STOP. 558 */ 559 qemu_mutex_lock(&s->free_page_lock); 560 /* 561 * The guest hasn't done the reporting, so host sends a notification 562 * to the guest to actively stop the reporting. 563 */ 564 s->free_page_report_status = FREE_PAGE_REPORT_S_STOP; 565 qemu_mutex_unlock(&s->free_page_lock); 566 virtio_notify_config(vdev); 567 } 568 } 569 570 static void virtio_balloon_free_page_done(VirtIOBalloon *s) 571 { 572 VirtIODevice *vdev = VIRTIO_DEVICE(s); 573 574 s->free_page_report_status = FREE_PAGE_REPORT_S_DONE; 575 virtio_notify_config(vdev); 576 } 577 578 static int 579 virtio_balloon_free_page_report_notify(NotifierWithReturn *n, void *data) 580 { 581 VirtIOBalloon *dev = container_of(n, VirtIOBalloon, 582 free_page_report_notify); 583 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 584 PrecopyNotifyData *pnd = data; 585 586 if (!virtio_balloon_free_page_support(dev)) { 587 /* 588 * This is an optimization provided to migration, so just return 0 to 589 * have the normal migration process not affected when this feature is 590 * not supported. 591 */ 592 return 0; 593 } 594 595 switch (pnd->reason) { 596 case PRECOPY_NOTIFY_SETUP: 597 precopy_enable_free_page_optimization(); 598 break; 599 case PRECOPY_NOTIFY_COMPLETE: 600 case PRECOPY_NOTIFY_CLEANUP: 601 case PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC: 602 virtio_balloon_free_page_stop(dev); 603 break; 604 case PRECOPY_NOTIFY_AFTER_BITMAP_SYNC: 605 if (vdev->vm_running) { 606 virtio_balloon_free_page_start(dev); 607 } else { 608 virtio_balloon_free_page_done(dev); 609 } 610 break; 611 default: 612 virtio_error(vdev, "%s: %d reason unknown", __func__, pnd->reason); 613 } 614 615 return 0; 616 } 617 618 static size_t virtio_balloon_config_size(VirtIOBalloon *s) 619 { 620 uint64_t features = s->host_features; 621 622 if (s->qemu_4_0_config_size) { 623 return sizeof(struct virtio_balloon_config); 624 } 625 if (virtio_has_feature(features, VIRTIO_BALLOON_F_PAGE_POISON)) { 626 return sizeof(struct virtio_balloon_config); 627 } 628 if (virtio_has_feature(features, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { 629 return offsetof(struct virtio_balloon_config, poison_val); 630 } 631 return offsetof(struct virtio_balloon_config, free_page_report_cmd_id); 632 } 633 634 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data) 635 { 636 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 637 struct virtio_balloon_config config = {}; 638 639 config.num_pages = cpu_to_le32(dev->num_pages); 640 config.actual = cpu_to_le32(dev->actual); 641 642 if (dev->free_page_report_status == FREE_PAGE_REPORT_S_REQUESTED) { 643 config.free_page_report_cmd_id = 644 cpu_to_le32(dev->free_page_report_cmd_id); 645 } else if (dev->free_page_report_status == FREE_PAGE_REPORT_S_STOP) { 646 config.free_page_report_cmd_id = 647 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_STOP); 648 } else if (dev->free_page_report_status == FREE_PAGE_REPORT_S_DONE) { 649 config.free_page_report_cmd_id = 650 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_DONE); 651 } 652 653 trace_virtio_balloon_get_config(config.num_pages, config.actual); 654 memcpy(config_data, &config, virtio_balloon_config_size(dev)); 655 } 656 657 static int build_dimm_list(Object *obj, void *opaque) 658 { 659 GSList **list = opaque; 660 661 if (object_dynamic_cast(obj, TYPE_PC_DIMM)) { 662 DeviceState *dev = DEVICE(obj); 663 if (dev->realized) { /* only realized DIMMs matter */ 664 *list = g_slist_prepend(*list, dev); 665 } 666 } 667 668 object_child_foreach(obj, build_dimm_list, opaque); 669 return 0; 670 } 671 672 static ram_addr_t get_current_ram_size(void) 673 { 674 GSList *list = NULL, *item; 675 ram_addr_t size = ram_size; 676 677 build_dimm_list(qdev_get_machine(), &list); 678 for (item = list; item; item = g_slist_next(item)) { 679 Object *obj = OBJECT(item->data); 680 if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) { 681 size += object_property_get_int(obj, PC_DIMM_SIZE_PROP, 682 &error_abort); 683 } 684 } 685 g_slist_free(list); 686 687 return size; 688 } 689 690 static void virtio_balloon_set_config(VirtIODevice *vdev, 691 const uint8_t *config_data) 692 { 693 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 694 struct virtio_balloon_config config; 695 uint32_t oldactual = dev->actual; 696 ram_addr_t vm_ram_size = get_current_ram_size(); 697 698 memcpy(&config, config_data, virtio_balloon_config_size(dev)); 699 dev->actual = le32_to_cpu(config.actual); 700 if (dev->actual != oldactual) { 701 qapi_event_send_balloon_change(vm_ram_size - 702 ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT)); 703 } 704 trace_virtio_balloon_set_config(dev->actual, oldactual); 705 } 706 707 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f, 708 Error **errp) 709 { 710 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 711 f |= dev->host_features; 712 virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ); 713 714 return f; 715 } 716 717 static void virtio_balloon_stat(void *opaque, BalloonInfo *info) 718 { 719 VirtIOBalloon *dev = opaque; 720 info->actual = get_current_ram_size() - ((uint64_t) dev->actual << 721 VIRTIO_BALLOON_PFN_SHIFT); 722 } 723 724 static void virtio_balloon_to_target(void *opaque, ram_addr_t target) 725 { 726 VirtIOBalloon *dev = VIRTIO_BALLOON(opaque); 727 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 728 ram_addr_t vm_ram_size = get_current_ram_size(); 729 730 if (target > vm_ram_size) { 731 target = vm_ram_size; 732 } 733 if (target) { 734 dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT; 735 virtio_notify_config(vdev); 736 } 737 trace_virtio_balloon_to_target(target, dev->num_pages); 738 } 739 740 static int virtio_balloon_post_load_device(void *opaque, int version_id) 741 { 742 VirtIOBalloon *s = VIRTIO_BALLOON(opaque); 743 744 if (balloon_stats_enabled(s)) { 745 balloon_stats_change_timer(s, s->stats_poll_interval); 746 } 747 return 0; 748 } 749 750 static const VMStateDescription vmstate_virtio_balloon_free_page_report = { 751 .name = "virtio-balloon-device/free-page-report", 752 .version_id = 1, 753 .minimum_version_id = 1, 754 .needed = virtio_balloon_free_page_support, 755 .fields = (VMStateField[]) { 756 VMSTATE_UINT32(free_page_report_cmd_id, VirtIOBalloon), 757 VMSTATE_UINT32(free_page_report_status, VirtIOBalloon), 758 VMSTATE_END_OF_LIST() 759 } 760 }; 761 762 static const VMStateDescription vmstate_virtio_balloon_device = { 763 .name = "virtio-balloon-device", 764 .version_id = 1, 765 .minimum_version_id = 1, 766 .post_load = virtio_balloon_post_load_device, 767 .fields = (VMStateField[]) { 768 VMSTATE_UINT32(num_pages, VirtIOBalloon), 769 VMSTATE_UINT32(actual, VirtIOBalloon), 770 VMSTATE_END_OF_LIST() 771 }, 772 .subsections = (const VMStateDescription * []) { 773 &vmstate_virtio_balloon_free_page_report, 774 NULL 775 } 776 }; 777 778 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp) 779 { 780 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 781 VirtIOBalloon *s = VIRTIO_BALLOON(dev); 782 int ret; 783 784 virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON, 785 virtio_balloon_config_size(s)); 786 787 ret = qemu_add_balloon_handler(virtio_balloon_to_target, 788 virtio_balloon_stat, s); 789 790 if (ret < 0) { 791 error_setg(errp, "Only one balloon device is supported"); 792 virtio_cleanup(vdev); 793 return; 794 } 795 796 s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output); 797 s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output); 798 s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats); 799 800 if (virtio_has_feature(s->host_features, 801 VIRTIO_BALLOON_F_FREE_PAGE_HINT)) { 802 s->free_page_vq = virtio_add_queue(vdev, VIRTQUEUE_MAX_SIZE, 803 virtio_balloon_handle_free_page_vq); 804 s->free_page_report_status = FREE_PAGE_REPORT_S_STOP; 805 s->free_page_report_cmd_id = 806 VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN; 807 s->free_page_report_notify.notify = 808 virtio_balloon_free_page_report_notify; 809 precopy_add_notifier(&s->free_page_report_notify); 810 if (s->iothread) { 811 object_ref(OBJECT(s->iothread)); 812 s->free_page_bh = aio_bh_new(iothread_get_aio_context(s->iothread), 813 virtio_ballloon_get_free_page_hints, s); 814 qemu_mutex_init(&s->free_page_lock); 815 qemu_cond_init(&s->free_page_cond); 816 s->block_iothread = false; 817 } else { 818 /* Simply disable this feature if the iothread wasn't created. */ 819 s->host_features &= ~(1 << VIRTIO_BALLOON_F_FREE_PAGE_HINT); 820 virtio_error(vdev, "iothread is missing"); 821 } 822 } 823 reset_stats(s); 824 } 825 826 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp) 827 { 828 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 829 VirtIOBalloon *s = VIRTIO_BALLOON(dev); 830 831 if (virtio_balloon_free_page_support(s)) { 832 qemu_bh_delete(s->free_page_bh); 833 virtio_balloon_free_page_stop(s); 834 precopy_remove_notifier(&s->free_page_report_notify); 835 } 836 balloon_stats_destroy_timer(s); 837 qemu_remove_balloon_handler(s); 838 virtio_cleanup(vdev); 839 } 840 841 static void virtio_balloon_device_reset(VirtIODevice *vdev) 842 { 843 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 844 845 if (virtio_balloon_free_page_support(s)) { 846 virtio_balloon_free_page_stop(s); 847 } 848 849 if (s->stats_vq_elem != NULL) { 850 virtqueue_unpop(s->svq, s->stats_vq_elem, 0); 851 g_free(s->stats_vq_elem); 852 s->stats_vq_elem = NULL; 853 } 854 } 855 856 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status) 857 { 858 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 859 860 if (!s->stats_vq_elem && vdev->vm_running && 861 (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) { 862 /* poll stats queue for the element we have discarded when the VM 863 * was stopped */ 864 virtio_balloon_receive_stats(vdev, s->svq); 865 } 866 867 if (virtio_balloon_free_page_support(s)) { 868 /* 869 * The VM is woken up and the iothread was blocked, so signal it to 870 * continue. 871 */ 872 if (vdev->vm_running && s->block_iothread) { 873 qemu_mutex_lock(&s->free_page_lock); 874 s->block_iothread = false; 875 qemu_cond_signal(&s->free_page_cond); 876 qemu_mutex_unlock(&s->free_page_lock); 877 } 878 879 /* The VM is stopped, block the iothread. */ 880 if (!vdev->vm_running) { 881 qemu_mutex_lock(&s->free_page_lock); 882 s->block_iothread = true; 883 qemu_mutex_unlock(&s->free_page_lock); 884 } 885 } 886 } 887 888 static void virtio_balloon_instance_init(Object *obj) 889 { 890 VirtIOBalloon *s = VIRTIO_BALLOON(obj); 891 892 object_property_add(obj, "guest-stats", "guest statistics", 893 balloon_stats_get_all, NULL, NULL, s, NULL); 894 895 object_property_add(obj, "guest-stats-polling-interval", "int", 896 balloon_stats_get_poll_interval, 897 balloon_stats_set_poll_interval, 898 NULL, s, NULL); 899 } 900 901 static const VMStateDescription vmstate_virtio_balloon = { 902 .name = "virtio-balloon", 903 .minimum_version_id = 1, 904 .version_id = 1, 905 .fields = (VMStateField[]) { 906 VMSTATE_VIRTIO_DEVICE, 907 VMSTATE_END_OF_LIST() 908 }, 909 }; 910 911 static Property virtio_balloon_properties[] = { 912 DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features, 913 VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false), 914 DEFINE_PROP_BIT("free-page-hint", VirtIOBalloon, host_features, 915 VIRTIO_BALLOON_F_FREE_PAGE_HINT, false), 916 /* QEMU 4.0 accidentally changed the config size even when free-page-hint 917 * is disabled, resulting in QEMU 3.1 migration incompatibility. This 918 * property retains this quirk for QEMU 4.1 machine types. 919 */ 920 DEFINE_PROP_BOOL("qemu-4-0-config-size", VirtIOBalloon, 921 qemu_4_0_config_size, false), 922 DEFINE_PROP_LINK("iothread", VirtIOBalloon, iothread, TYPE_IOTHREAD, 923 IOThread *), 924 DEFINE_PROP_END_OF_LIST(), 925 }; 926 927 static void virtio_balloon_class_init(ObjectClass *klass, void *data) 928 { 929 DeviceClass *dc = DEVICE_CLASS(klass); 930 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 931 932 dc->props = virtio_balloon_properties; 933 dc->vmsd = &vmstate_virtio_balloon; 934 set_bit(DEVICE_CATEGORY_MISC, dc->categories); 935 vdc->realize = virtio_balloon_device_realize; 936 vdc->unrealize = virtio_balloon_device_unrealize; 937 vdc->reset = virtio_balloon_device_reset; 938 vdc->get_config = virtio_balloon_get_config; 939 vdc->set_config = virtio_balloon_set_config; 940 vdc->get_features = virtio_balloon_get_features; 941 vdc->set_status = virtio_balloon_set_status; 942 vdc->vmsd = &vmstate_virtio_balloon_device; 943 } 944 945 static const TypeInfo virtio_balloon_info = { 946 .name = TYPE_VIRTIO_BALLOON, 947 .parent = TYPE_VIRTIO_DEVICE, 948 .instance_size = sizeof(VirtIOBalloon), 949 .instance_init = virtio_balloon_instance_init, 950 .class_init = virtio_balloon_class_init, 951 }; 952 953 static void virtio_register_types(void) 954 { 955 type_register_static(&virtio_balloon_info); 956 } 957 958 type_init(virtio_register_types) 959