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/timer.h" 19 #include "qemu-common.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 31 #include "hw/virtio/virtio-bus.h" 32 #include "hw/virtio/virtio-access.h" 33 34 #define BALLOON_PAGE_SIZE (1 << VIRTIO_BALLOON_PFN_SHIFT) 35 36 struct PartiallyBalloonedPage { 37 RAMBlock *rb; 38 ram_addr_t base; 39 unsigned long bitmap[]; 40 }; 41 42 static void balloon_inflate_page(VirtIOBalloon *balloon, 43 MemoryRegion *mr, hwaddr offset) 44 { 45 void *addr = memory_region_get_ram_ptr(mr) + offset; 46 RAMBlock *rb; 47 size_t rb_page_size; 48 int subpages; 49 ram_addr_t ram_offset, host_page_base; 50 51 /* XXX is there a better way to get to the RAMBlock than via a 52 * host address? */ 53 rb = qemu_ram_block_from_host(addr, false, &ram_offset); 54 rb_page_size = qemu_ram_pagesize(rb); 55 host_page_base = ram_offset & ~(rb_page_size - 1); 56 57 if (rb_page_size == BALLOON_PAGE_SIZE) { 58 /* Easy case */ 59 60 ram_block_discard_range(rb, ram_offset, rb_page_size); 61 /* We ignore errors from ram_block_discard_range(), because it 62 * has already reported them, and failing to discard a balloon 63 * page is not fatal */ 64 return; 65 } 66 67 /* Hard case 68 * 69 * We've put a piece of a larger host page into the balloon - we 70 * need to keep track until we have a whole host page to 71 * discard 72 */ 73 warn_report_once( 74 "Balloon used with backing page size > 4kiB, this may not be reliable"); 75 76 subpages = rb_page_size / BALLOON_PAGE_SIZE; 77 78 if (balloon->pbp 79 && (rb != balloon->pbp->rb 80 || host_page_base != balloon->pbp->base)) { 81 /* We've partially ballooned part of a host page, but now 82 * we're trying to balloon part of a different one. Too hard, 83 * give up on the old partial page */ 84 free(balloon->pbp); 85 balloon->pbp = NULL; 86 } 87 88 if (!balloon->pbp) { 89 /* Starting on a new host page */ 90 size_t bitlen = BITS_TO_LONGS(subpages) * sizeof(unsigned long); 91 balloon->pbp = g_malloc0(sizeof(PartiallyBalloonedPage) + bitlen); 92 balloon->pbp->rb = rb; 93 balloon->pbp->base = host_page_base; 94 } 95 96 bitmap_set(balloon->pbp->bitmap, 97 (ram_offset - balloon->pbp->base) / BALLOON_PAGE_SIZE, 98 subpages); 99 100 if (bitmap_full(balloon->pbp->bitmap, subpages)) { 101 /* We've accumulated a full host page, we can actually discard 102 * it now */ 103 104 ram_block_discard_range(rb, balloon->pbp->base, rb_page_size); 105 /* We ignore errors from ram_block_discard_range(), because it 106 * has already reported them, and failing to discard a balloon 107 * page is not fatal */ 108 109 free(balloon->pbp); 110 balloon->pbp = NULL; 111 } 112 } 113 114 static const char *balloon_stat_names[] = { 115 [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in", 116 [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out", 117 [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults", 118 [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults", 119 [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory", 120 [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory", 121 [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory", 122 [VIRTIO_BALLOON_S_CACHES] = "stat-disk-caches", 123 [VIRTIO_BALLOON_S_HTLB_PGALLOC] = "stat-htlb-pgalloc", 124 [VIRTIO_BALLOON_S_HTLB_PGFAIL] = "stat-htlb-pgfail", 125 [VIRTIO_BALLOON_S_NR] = NULL 126 }; 127 128 /* 129 * reset_stats - Mark all items in the stats array as unset 130 * 131 * This function needs to be called at device initialization and before 132 * updating to a set of newly-generated stats. This will ensure that no 133 * stale values stick around in case the guest reports a subset of the supported 134 * statistics. 135 */ 136 static inline void reset_stats(VirtIOBalloon *dev) 137 { 138 int i; 139 for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1); 140 } 141 142 static bool balloon_stats_supported(const VirtIOBalloon *s) 143 { 144 VirtIODevice *vdev = VIRTIO_DEVICE(s); 145 return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ); 146 } 147 148 static bool balloon_stats_enabled(const VirtIOBalloon *s) 149 { 150 return s->stats_poll_interval > 0; 151 } 152 153 static void balloon_stats_destroy_timer(VirtIOBalloon *s) 154 { 155 if (balloon_stats_enabled(s)) { 156 timer_del(s->stats_timer); 157 timer_free(s->stats_timer); 158 s->stats_timer = NULL; 159 s->stats_poll_interval = 0; 160 } 161 } 162 163 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs) 164 { 165 timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000); 166 } 167 168 static void balloon_stats_poll_cb(void *opaque) 169 { 170 VirtIOBalloon *s = opaque; 171 VirtIODevice *vdev = VIRTIO_DEVICE(s); 172 173 if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) { 174 /* re-schedule */ 175 balloon_stats_change_timer(s, s->stats_poll_interval); 176 return; 177 } 178 179 virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset); 180 virtio_notify(vdev, s->svq); 181 g_free(s->stats_vq_elem); 182 s->stats_vq_elem = NULL; 183 } 184 185 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name, 186 void *opaque, Error **errp) 187 { 188 Error *err = NULL; 189 VirtIOBalloon *s = opaque; 190 int i; 191 192 visit_start_struct(v, name, NULL, 0, &err); 193 if (err) { 194 goto out; 195 } 196 visit_type_int(v, "last-update", &s->stats_last_update, &err); 197 if (err) { 198 goto out_end; 199 } 200 201 visit_start_struct(v, "stats", NULL, 0, &err); 202 if (err) { 203 goto out_end; 204 } 205 for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) { 206 visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err); 207 if (err) { 208 goto out_nested; 209 } 210 } 211 visit_check_struct(v, &err); 212 out_nested: 213 visit_end_struct(v, NULL); 214 215 if (!err) { 216 visit_check_struct(v, &err); 217 } 218 out_end: 219 visit_end_struct(v, NULL); 220 out: 221 error_propagate(errp, err); 222 } 223 224 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v, 225 const char *name, void *opaque, 226 Error **errp) 227 { 228 VirtIOBalloon *s = opaque; 229 visit_type_int(v, name, &s->stats_poll_interval, errp); 230 } 231 232 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v, 233 const char *name, void *opaque, 234 Error **errp) 235 { 236 VirtIOBalloon *s = opaque; 237 Error *local_err = NULL; 238 int64_t value; 239 240 visit_type_int(v, name, &value, &local_err); 241 if (local_err) { 242 error_propagate(errp, local_err); 243 return; 244 } 245 246 if (value < 0) { 247 error_setg(errp, "timer value must be greater than zero"); 248 return; 249 } 250 251 if (value > UINT32_MAX) { 252 error_setg(errp, "timer value is too big"); 253 return; 254 } 255 256 if (value == s->stats_poll_interval) { 257 return; 258 } 259 260 if (value == 0) { 261 /* timer=0 disables the timer */ 262 balloon_stats_destroy_timer(s); 263 return; 264 } 265 266 if (balloon_stats_enabled(s)) { 267 /* timer interval change */ 268 s->stats_poll_interval = value; 269 balloon_stats_change_timer(s, value); 270 return; 271 } 272 273 /* create a new timer */ 274 g_assert(s->stats_timer == NULL); 275 s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s); 276 s->stats_poll_interval = value; 277 balloon_stats_change_timer(s, 0); 278 } 279 280 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq) 281 { 282 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 283 VirtQueueElement *elem; 284 MemoryRegionSection section; 285 286 for (;;) { 287 size_t offset = 0; 288 uint32_t pfn; 289 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 290 if (!elem) { 291 return; 292 } 293 294 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) { 295 hwaddr pa; 296 int p = virtio_ldl_p(vdev, &pfn); 297 298 pa = (hwaddr) p << VIRTIO_BALLOON_PFN_SHIFT; 299 offset += 4; 300 301 section = memory_region_find(get_system_memory(), pa, 302 BALLOON_PAGE_SIZE); 303 if (!section.mr) { 304 trace_virtio_balloon_bad_addr(pa); 305 continue; 306 } 307 if (!memory_region_is_ram(section.mr) || 308 memory_region_is_rom(section.mr) || 309 memory_region_is_romd(section.mr)) { 310 trace_virtio_balloon_bad_addr(pa); 311 memory_region_unref(section.mr); 312 continue; 313 } 314 315 trace_virtio_balloon_handle_output(memory_region_name(section.mr), 316 pa); 317 if (!qemu_balloon_is_inhibited() && vq != s->dvq) { 318 balloon_inflate_page(s, section.mr, section.offset_within_region); 319 } 320 memory_region_unref(section.mr); 321 } 322 323 virtqueue_push(vq, elem, offset); 324 virtio_notify(vdev, vq); 325 g_free(elem); 326 } 327 } 328 329 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq) 330 { 331 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 332 VirtQueueElement *elem; 333 VirtIOBalloonStat stat; 334 size_t offset = 0; 335 qemu_timeval tv; 336 337 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 338 if (!elem) { 339 goto out; 340 } 341 342 if (s->stats_vq_elem != NULL) { 343 /* This should never happen if the driver follows the spec. */ 344 virtqueue_push(vq, s->stats_vq_elem, 0); 345 virtio_notify(vdev, vq); 346 g_free(s->stats_vq_elem); 347 } 348 349 s->stats_vq_elem = elem; 350 351 /* Initialize the stats to get rid of any stale values. This is only 352 * needed to handle the case where a guest supports fewer stats than it 353 * used to (ie. it has booted into an old kernel). 354 */ 355 reset_stats(s); 356 357 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat)) 358 == sizeof(stat)) { 359 uint16_t tag = virtio_tswap16(vdev, stat.tag); 360 uint64_t val = virtio_tswap64(vdev, stat.val); 361 362 offset += sizeof(stat); 363 if (tag < VIRTIO_BALLOON_S_NR) 364 s->stats[tag] = val; 365 } 366 s->stats_vq_offset = offset; 367 368 if (qemu_gettimeofday(&tv) < 0) { 369 warn_report("%s: failed to get time of day", __func__); 370 goto out; 371 } 372 373 s->stats_last_update = tv.tv_sec; 374 375 out: 376 if (balloon_stats_enabled(s)) { 377 balloon_stats_change_timer(s, s->stats_poll_interval); 378 } 379 } 380 381 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data) 382 { 383 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 384 struct virtio_balloon_config config = {}; 385 386 config.num_pages = cpu_to_le32(dev->num_pages); 387 config.actual = cpu_to_le32(dev->actual); 388 389 trace_virtio_balloon_get_config(config.num_pages, config.actual); 390 memcpy(config_data, &config, sizeof(struct virtio_balloon_config)); 391 } 392 393 static int build_dimm_list(Object *obj, void *opaque) 394 { 395 GSList **list = opaque; 396 397 if (object_dynamic_cast(obj, TYPE_PC_DIMM)) { 398 DeviceState *dev = DEVICE(obj); 399 if (dev->realized) { /* only realized DIMMs matter */ 400 *list = g_slist_prepend(*list, dev); 401 } 402 } 403 404 object_child_foreach(obj, build_dimm_list, opaque); 405 return 0; 406 } 407 408 static ram_addr_t get_current_ram_size(void) 409 { 410 GSList *list = NULL, *item; 411 ram_addr_t size = ram_size; 412 413 build_dimm_list(qdev_get_machine(), &list); 414 for (item = list; item; item = g_slist_next(item)) { 415 Object *obj = OBJECT(item->data); 416 if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) { 417 size += object_property_get_int(obj, PC_DIMM_SIZE_PROP, 418 &error_abort); 419 } 420 } 421 g_slist_free(list); 422 423 return size; 424 } 425 426 static void virtio_balloon_set_config(VirtIODevice *vdev, 427 const uint8_t *config_data) 428 { 429 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 430 struct virtio_balloon_config config; 431 uint32_t oldactual = dev->actual; 432 ram_addr_t vm_ram_size = get_current_ram_size(); 433 434 memcpy(&config, config_data, sizeof(struct virtio_balloon_config)); 435 dev->actual = le32_to_cpu(config.actual); 436 if (dev->actual != oldactual) { 437 qapi_event_send_balloon_change(vm_ram_size - 438 ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT)); 439 } 440 trace_virtio_balloon_set_config(dev->actual, oldactual); 441 } 442 443 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f, 444 Error **errp) 445 { 446 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); 447 f |= dev->host_features; 448 virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ); 449 return f; 450 } 451 452 static void virtio_balloon_stat(void *opaque, BalloonInfo *info) 453 { 454 VirtIOBalloon *dev = opaque; 455 info->actual = get_current_ram_size() - ((uint64_t) dev->actual << 456 VIRTIO_BALLOON_PFN_SHIFT); 457 } 458 459 static void virtio_balloon_to_target(void *opaque, ram_addr_t target) 460 { 461 VirtIOBalloon *dev = VIRTIO_BALLOON(opaque); 462 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 463 ram_addr_t vm_ram_size = get_current_ram_size(); 464 465 if (target > vm_ram_size) { 466 target = vm_ram_size; 467 } 468 if (target) { 469 dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT; 470 virtio_notify_config(vdev); 471 } 472 trace_virtio_balloon_to_target(target, dev->num_pages); 473 } 474 475 static int virtio_balloon_post_load_device(void *opaque, int version_id) 476 { 477 VirtIOBalloon *s = VIRTIO_BALLOON(opaque); 478 479 if (balloon_stats_enabled(s)) { 480 balloon_stats_change_timer(s, s->stats_poll_interval); 481 } 482 return 0; 483 } 484 485 static const VMStateDescription vmstate_virtio_balloon_device = { 486 .name = "virtio-balloon-device", 487 .version_id = 1, 488 .minimum_version_id = 1, 489 .post_load = virtio_balloon_post_load_device, 490 .fields = (VMStateField[]) { 491 VMSTATE_UINT32(num_pages, VirtIOBalloon), 492 VMSTATE_UINT32(actual, VirtIOBalloon), 493 VMSTATE_END_OF_LIST() 494 }, 495 }; 496 497 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp) 498 { 499 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 500 VirtIOBalloon *s = VIRTIO_BALLOON(dev); 501 int ret; 502 503 virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON, 504 sizeof(struct virtio_balloon_config)); 505 506 ret = qemu_add_balloon_handler(virtio_balloon_to_target, 507 virtio_balloon_stat, s); 508 509 if (ret < 0) { 510 error_setg(errp, "Only one balloon device is supported"); 511 virtio_cleanup(vdev); 512 return; 513 } 514 515 s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output); 516 s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output); 517 s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats); 518 519 reset_stats(s); 520 } 521 522 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp) 523 { 524 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 525 VirtIOBalloon *s = VIRTIO_BALLOON(dev); 526 527 balloon_stats_destroy_timer(s); 528 qemu_remove_balloon_handler(s); 529 virtio_cleanup(vdev); 530 } 531 532 static void virtio_balloon_device_reset(VirtIODevice *vdev) 533 { 534 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 535 536 if (s->stats_vq_elem != NULL) { 537 virtqueue_unpop(s->svq, s->stats_vq_elem, 0); 538 g_free(s->stats_vq_elem); 539 s->stats_vq_elem = NULL; 540 } 541 } 542 543 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status) 544 { 545 VirtIOBalloon *s = VIRTIO_BALLOON(vdev); 546 547 if (!s->stats_vq_elem && vdev->vm_running && 548 (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) { 549 /* poll stats queue for the element we have discarded when the VM 550 * was stopped */ 551 virtio_balloon_receive_stats(vdev, s->svq); 552 } 553 } 554 555 static void virtio_balloon_instance_init(Object *obj) 556 { 557 VirtIOBalloon *s = VIRTIO_BALLOON(obj); 558 559 object_property_add(obj, "guest-stats", "guest statistics", 560 balloon_stats_get_all, NULL, NULL, s, NULL); 561 562 object_property_add(obj, "guest-stats-polling-interval", "int", 563 balloon_stats_get_poll_interval, 564 balloon_stats_set_poll_interval, 565 NULL, s, NULL); 566 } 567 568 static const VMStateDescription vmstate_virtio_balloon = { 569 .name = "virtio-balloon", 570 .minimum_version_id = 1, 571 .version_id = 1, 572 .fields = (VMStateField[]) { 573 VMSTATE_VIRTIO_DEVICE, 574 VMSTATE_END_OF_LIST() 575 }, 576 }; 577 578 static Property virtio_balloon_properties[] = { 579 DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features, 580 VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false), 581 DEFINE_PROP_END_OF_LIST(), 582 }; 583 584 static void virtio_balloon_class_init(ObjectClass *klass, void *data) 585 { 586 DeviceClass *dc = DEVICE_CLASS(klass); 587 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 588 589 dc->props = virtio_balloon_properties; 590 dc->vmsd = &vmstate_virtio_balloon; 591 set_bit(DEVICE_CATEGORY_MISC, dc->categories); 592 vdc->realize = virtio_balloon_device_realize; 593 vdc->unrealize = virtio_balloon_device_unrealize; 594 vdc->reset = virtio_balloon_device_reset; 595 vdc->get_config = virtio_balloon_get_config; 596 vdc->set_config = virtio_balloon_set_config; 597 vdc->get_features = virtio_balloon_get_features; 598 vdc->set_status = virtio_balloon_set_status; 599 vdc->vmsd = &vmstate_virtio_balloon_device; 600 } 601 602 static const TypeInfo virtio_balloon_info = { 603 .name = TYPE_VIRTIO_BALLOON, 604 .parent = TYPE_VIRTIO_DEVICE, 605 .instance_size = sizeof(VirtIOBalloon), 606 .instance_init = virtio_balloon_instance_init, 607 .class_init = virtio_balloon_class_init, 608 }; 609 610 static void virtio_register_types(void) 611 { 612 type_register_static(&virtio_balloon_info); 613 } 614 615 type_init(virtio_register_types) 616