qemu/hw/virtio/virtio-balloon.c
<<
>>
Prefs
   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/i386/pc.h"
  22#include "sysemu/balloon.h"
  23#include "hw/virtio/virtio-balloon.h"
  24#include "sysemu/kvm.h"
  25#include "exec/address-spaces.h"
  26#include "qapi/visitor.h"
  27#include "qapi-event.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
  36static void balloon_page(void *addr, int deflate)
  37{
  38    if (!qemu_balloon_is_inhibited() && (!kvm_enabled() ||
  39                                         kvm_has_sync_mmu())) {
  40        qemu_madvise(addr, BALLOON_PAGE_SIZE,
  41                deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
  42    }
  43}
  44
  45static const char *balloon_stat_names[] = {
  46   [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in",
  47   [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out",
  48   [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults",
  49   [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults",
  50   [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory",
  51   [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory",
  52   [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory",
  53   [VIRTIO_BALLOON_S_NR] = NULL
  54};
  55
  56/*
  57 * reset_stats - Mark all items in the stats array as unset
  58 *
  59 * This function needs to be called at device initialization and before
  60 * updating to a set of newly-generated stats.  This will ensure that no
  61 * stale values stick around in case the guest reports a subset of the supported
  62 * statistics.
  63 */
  64static inline void reset_stats(VirtIOBalloon *dev)
  65{
  66    int i;
  67    for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
  68}
  69
  70static bool balloon_stats_supported(const VirtIOBalloon *s)
  71{
  72    VirtIODevice *vdev = VIRTIO_DEVICE(s);
  73    return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ);
  74}
  75
  76static bool balloon_stats_enabled(const VirtIOBalloon *s)
  77{
  78    return s->stats_poll_interval > 0;
  79}
  80
  81static void balloon_stats_destroy_timer(VirtIOBalloon *s)
  82{
  83    if (balloon_stats_enabled(s)) {
  84        timer_del(s->stats_timer);
  85        timer_free(s->stats_timer);
  86        s->stats_timer = NULL;
  87        s->stats_poll_interval = 0;
  88    }
  89}
  90
  91static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs)
  92{
  93    timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000);
  94}
  95
  96static void balloon_stats_poll_cb(void *opaque)
  97{
  98    VirtIOBalloon *s = opaque;
  99    VirtIODevice *vdev = VIRTIO_DEVICE(s);
 100
 101    if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) {
 102        /* re-schedule */
 103        balloon_stats_change_timer(s, s->stats_poll_interval);
 104        return;
 105    }
 106
 107    virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset);
 108    virtio_notify(vdev, s->svq);
 109    g_free(s->stats_vq_elem);
 110    s->stats_vq_elem = NULL;
 111}
 112
 113static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name,
 114                                  void *opaque, Error **errp)
 115{
 116    Error *err = NULL;
 117    VirtIOBalloon *s = opaque;
 118    int i;
 119
 120    visit_start_struct(v, name, NULL, 0, &err);
 121    if (err) {
 122        goto out;
 123    }
 124    visit_type_int(v, "last-update", &s->stats_last_update, &err);
 125    if (err) {
 126        goto out_end;
 127    }
 128
 129    visit_start_struct(v, "stats", NULL, 0, &err);
 130    if (err) {
 131        goto out_end;
 132    }
 133    for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) {
 134        visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err);
 135        if (err) {
 136            goto out_nested;
 137        }
 138    }
 139    visit_check_struct(v, &err);
 140out_nested:
 141    visit_end_struct(v, NULL);
 142
 143    if (!err) {
 144        visit_check_struct(v, &err);
 145    }
 146out_end:
 147    visit_end_struct(v, NULL);
 148out:
 149    error_propagate(errp, err);
 150}
 151
 152static void balloon_stats_get_poll_interval(Object *obj, Visitor *v,
 153                                            const char *name, void *opaque,
 154                                            Error **errp)
 155{
 156    VirtIOBalloon *s = opaque;
 157    visit_type_int(v, name, &s->stats_poll_interval, errp);
 158}
 159
 160static void balloon_stats_set_poll_interval(Object *obj, Visitor *v,
 161                                            const char *name, void *opaque,
 162                                            Error **errp)
 163{
 164    VirtIOBalloon *s = opaque;
 165    Error *local_err = NULL;
 166    int64_t value;
 167
 168    visit_type_int(v, name, &value, &local_err);
 169    if (local_err) {
 170        error_propagate(errp, local_err);
 171        return;
 172    }
 173
 174    if (value < 0) {
 175        error_setg(errp, "timer value must be greater than zero");
 176        return;
 177    }
 178
 179    if (value > UINT32_MAX) {
 180        error_setg(errp, "timer value is too big");
 181        return;
 182    }
 183
 184    if (value == s->stats_poll_interval) {
 185        return;
 186    }
 187
 188    if (value == 0) {
 189        /* timer=0 disables the timer */
 190        balloon_stats_destroy_timer(s);
 191        return;
 192    }
 193
 194    if (balloon_stats_enabled(s)) {
 195        /* timer interval change */
 196        s->stats_poll_interval = value;
 197        balloon_stats_change_timer(s, value);
 198        return;
 199    }
 200
 201    /* create a new timer */
 202    g_assert(s->stats_timer == NULL);
 203    s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s);
 204    s->stats_poll_interval = value;
 205    balloon_stats_change_timer(s, 0);
 206}
 207
 208static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 209{
 210    VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
 211    VirtQueueElement *elem;
 212    MemoryRegionSection section;
 213
 214    for (;;) {
 215        size_t offset = 0;
 216        uint32_t pfn;
 217        elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
 218        if (!elem) {
 219            return;
 220        }
 221
 222        while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) {
 223            ram_addr_t pa;
 224            ram_addr_t addr;
 225            int p = virtio_ldl_p(vdev, &pfn);
 226
 227            pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT;
 228            offset += 4;
 229
 230            /* FIXME: remove get_system_memory(), but how? */
 231            section = memory_region_find(get_system_memory(), pa, 1);
 232            if (!int128_nz(section.size) ||
 233                !memory_region_is_ram(section.mr) ||
 234                memory_region_is_rom(section.mr) ||
 235                memory_region_is_romd(section.mr)) {
 236                trace_virtio_balloon_bad_addr(pa);
 237                memory_region_unref(section.mr);
 238                continue;
 239            }
 240
 241            trace_virtio_balloon_handle_output(memory_region_name(section.mr),
 242                                               pa);
 243            /* Using memory_region_get_ram_ptr is bending the rules a bit, but
 244               should be OK because we only want a single page.  */
 245            addr = section.offset_within_region;
 246            balloon_page(memory_region_get_ram_ptr(section.mr) + addr,
 247                         !!(vq == s->dvq));
 248            memory_region_unref(section.mr);
 249        }
 250
 251        virtqueue_push(vq, elem, offset);
 252        virtio_notify(vdev, vq);
 253        g_free(elem);
 254    }
 255}
 256
 257static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
 258{
 259    VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
 260    VirtQueueElement *elem;
 261    VirtIOBalloonStat stat;
 262    size_t offset = 0;
 263    qemu_timeval tv;
 264
 265    elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
 266    if (!elem) {
 267        goto out;
 268    }
 269
 270    if (s->stats_vq_elem != NULL) {
 271        /* This should never happen if the driver follows the spec. */
 272        virtqueue_push(vq, s->stats_vq_elem, 0);
 273        virtio_notify(vdev, vq);
 274        g_free(s->stats_vq_elem);
 275    }
 276
 277    s->stats_vq_elem = elem;
 278
 279    /* Initialize the stats to get rid of any stale values.  This is only
 280     * needed to handle the case where a guest supports fewer stats than it
 281     * used to (ie. it has booted into an old kernel).
 282     */
 283    reset_stats(s);
 284
 285    while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat))
 286           == sizeof(stat)) {
 287        uint16_t tag = virtio_tswap16(vdev, stat.tag);
 288        uint64_t val = virtio_tswap64(vdev, stat.val);
 289
 290        offset += sizeof(stat);
 291        if (tag < VIRTIO_BALLOON_S_NR)
 292            s->stats[tag] = val;
 293    }
 294    s->stats_vq_offset = offset;
 295
 296    if (qemu_gettimeofday(&tv) < 0) {
 297        warn_report("%s: failed to get time of day", __func__);
 298        goto out;
 299    }
 300
 301    s->stats_last_update = tv.tv_sec;
 302
 303out:
 304    if (balloon_stats_enabled(s)) {
 305        balloon_stats_change_timer(s, s->stats_poll_interval);
 306    }
 307}
 308
 309static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
 310{
 311    VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
 312    struct virtio_balloon_config config;
 313
 314    config.num_pages = cpu_to_le32(dev->num_pages);
 315    config.actual = cpu_to_le32(dev->actual);
 316
 317    trace_virtio_balloon_get_config(config.num_pages, config.actual);
 318    memcpy(config_data, &config, sizeof(struct virtio_balloon_config));
 319}
 320
 321static int build_dimm_list(Object *obj, void *opaque)
 322{
 323    GSList **list = opaque;
 324
 325    if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 326        DeviceState *dev = DEVICE(obj);
 327        if (dev->realized) { /* only realized DIMMs matter */
 328            *list = g_slist_prepend(*list, dev);
 329        }
 330    }
 331
 332    object_child_foreach(obj, build_dimm_list, opaque);
 333    return 0;
 334}
 335
 336static ram_addr_t get_current_ram_size(void)
 337{
 338    GSList *list = NULL, *item;
 339    ram_addr_t size = ram_size;
 340
 341    build_dimm_list(qdev_get_machine(), &list);
 342    for (item = list; item; item = g_slist_next(item)) {
 343        Object *obj = OBJECT(item->data);
 344        if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) {
 345            size += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
 346                                            &error_abort);
 347        }
 348    }
 349    g_slist_free(list);
 350
 351    return size;
 352}
 353
 354static void virtio_balloon_set_config(VirtIODevice *vdev,
 355                                      const uint8_t *config_data)
 356{
 357    VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
 358    struct virtio_balloon_config config;
 359    uint32_t oldactual = dev->actual;
 360    ram_addr_t vm_ram_size = get_current_ram_size();
 361
 362    memcpy(&config, config_data, sizeof(struct virtio_balloon_config));
 363    dev->actual = le32_to_cpu(config.actual);
 364    if (dev->actual != oldactual) {
 365        qapi_event_send_balloon_change(vm_ram_size -
 366                        ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT),
 367                        &error_abort);
 368    }
 369    trace_virtio_balloon_set_config(dev->actual, oldactual);
 370}
 371
 372static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
 373                                            Error **errp)
 374{
 375    VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
 376    f |= dev->host_features;
 377    virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
 378    return f;
 379}
 380
 381static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
 382{
 383    VirtIOBalloon *dev = opaque;
 384    info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
 385                                             VIRTIO_BALLOON_PFN_SHIFT);
 386}
 387
 388static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
 389{
 390    VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
 391    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 392    ram_addr_t vm_ram_size = get_current_ram_size();
 393
 394    if (target > vm_ram_size) {
 395        target = vm_ram_size;
 396    }
 397    if (target) {
 398        dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
 399        virtio_notify_config(vdev);
 400    }
 401    trace_virtio_balloon_to_target(target, dev->num_pages);
 402}
 403
 404static int virtio_balloon_post_load_device(void *opaque, int version_id)
 405{
 406    VirtIOBalloon *s = VIRTIO_BALLOON(opaque);
 407
 408    if (balloon_stats_enabled(s)) {
 409        balloon_stats_change_timer(s, s->stats_poll_interval);
 410    }
 411    return 0;
 412}
 413
 414static const VMStateDescription vmstate_virtio_balloon_device = {
 415    .name = "virtio-balloon-device",
 416    .version_id = 1,
 417    .minimum_version_id = 1,
 418    .post_load = virtio_balloon_post_load_device,
 419    .fields = (VMStateField[]) {
 420        VMSTATE_UINT32(num_pages, VirtIOBalloon),
 421        VMSTATE_UINT32(actual, VirtIOBalloon),
 422        VMSTATE_END_OF_LIST()
 423    },
 424};
 425
 426static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
 427{
 428    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 429    VirtIOBalloon *s = VIRTIO_BALLOON(dev);
 430    int ret;
 431
 432    virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
 433                sizeof(struct virtio_balloon_config));
 434
 435    ret = qemu_add_balloon_handler(virtio_balloon_to_target,
 436                                   virtio_balloon_stat, s);
 437
 438    if (ret < 0) {
 439        error_setg(errp, "Only one balloon device is supported");
 440        virtio_cleanup(vdev);
 441        return;
 442    }
 443
 444    s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
 445    s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
 446    s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
 447
 448    reset_stats(s);
 449}
 450
 451static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp)
 452{
 453    VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 454    VirtIOBalloon *s = VIRTIO_BALLOON(dev);
 455
 456    balloon_stats_destroy_timer(s);
 457    qemu_remove_balloon_handler(s);
 458    virtio_cleanup(vdev);
 459}
 460
 461static void virtio_balloon_device_reset(VirtIODevice *vdev)
 462{
 463    VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
 464
 465    if (s->stats_vq_elem != NULL) {
 466        virtqueue_unpop(s->svq, s->stats_vq_elem, 0);
 467        g_free(s->stats_vq_elem);
 468        s->stats_vq_elem = NULL;
 469    }
 470}
 471
 472static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status)
 473{
 474    VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
 475
 476    if (!s->stats_vq_elem && vdev->vm_running &&
 477        (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) {
 478        /* poll stats queue for the element we have discarded when the VM
 479         * was stopped */
 480        virtio_balloon_receive_stats(vdev, s->svq);
 481    }
 482}
 483
 484static void virtio_balloon_instance_init(Object *obj)
 485{
 486    VirtIOBalloon *s = VIRTIO_BALLOON(obj);
 487
 488    object_property_add(obj, "guest-stats", "guest statistics",
 489                        balloon_stats_get_all, NULL, NULL, s, NULL);
 490
 491    object_property_add(obj, "guest-stats-polling-interval", "int",
 492                        balloon_stats_get_poll_interval,
 493                        balloon_stats_set_poll_interval,
 494                        NULL, s, NULL);
 495}
 496
 497static const VMStateDescription vmstate_virtio_balloon = {
 498    .name = "virtio-balloon",
 499    .minimum_version_id = 1,
 500    .version_id = 1,
 501    .fields = (VMStateField[]) {
 502        VMSTATE_VIRTIO_DEVICE,
 503        VMSTATE_END_OF_LIST()
 504    },
 505};
 506
 507static Property virtio_balloon_properties[] = {
 508    DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features,
 509                    VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false),
 510    DEFINE_PROP_END_OF_LIST(),
 511};
 512
 513static void virtio_balloon_class_init(ObjectClass *klass, void *data)
 514{
 515    DeviceClass *dc = DEVICE_CLASS(klass);
 516    VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
 517
 518    dc->props = virtio_balloon_properties;
 519    dc->vmsd = &vmstate_virtio_balloon;
 520    set_bit(DEVICE_CATEGORY_MISC, dc->categories);
 521    vdc->realize = virtio_balloon_device_realize;
 522    vdc->unrealize = virtio_balloon_device_unrealize;
 523    vdc->reset = virtio_balloon_device_reset;
 524    vdc->get_config = virtio_balloon_get_config;
 525    vdc->set_config = virtio_balloon_set_config;
 526    vdc->get_features = virtio_balloon_get_features;
 527    vdc->set_status = virtio_balloon_set_status;
 528    vdc->vmsd = &vmstate_virtio_balloon_device;
 529}
 530
 531static const TypeInfo virtio_balloon_info = {
 532    .name = TYPE_VIRTIO_BALLOON,
 533    .parent = TYPE_VIRTIO_DEVICE,
 534    .instance_size = sizeof(VirtIOBalloon),
 535    .instance_init = virtio_balloon_instance_init,
 536    .class_init = virtio_balloon_class_init,
 537};
 538
 539static void virtio_register_types(void)
 540{
 541    type_register_static(&virtio_balloon_info);
 542}
 543
 544type_init(virtio_register_types)
 545