hw/virtio-balloon.c

   1 /*
   2  * Virtio Block Device
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "iov.h"
  15 #include "qemu-common.h"
  16 #include "virtio.h"
  17 #include "pc.h"
  18 #include "sysemu.h"
  19 #include "cpu.h"
  20 #include "monitor.h"
  21 #include "balloon.h"
  22 #include "virtio-balloon.h"
  23 #include "kvm.h"
  24 #include "qemu-kvm.h"
  25 #include "qlist.h"
  26 #include "qint.h"
  27 #include "qstring.h"
  28
  29 #if defined(__linux__)
  30 #include <sys/mman.h>
  31 #endif
  32
  33 typedef struct VirtIOBalloon
  34 {
  35     VirtIODevice vdev;
  36     VirtQueue *ivq, *dvq, *svq;
  37     uint32_t num_pages;
  38     uint32_t actual;
  39     uint64_t stats[VIRTIO_BALLOON_S_NR];
  40     VirtQueueElement stats_vq_elem;
  41     size_t stats_vq_offset;
  42     MonitorCompletion *stats_callback;
  43     void *stats_opaque_callback_data;
  44 } VirtIOBalloon;
  45
  46 static VirtIOBalloon *to_virtio_balloon(VirtIODevice *vdev)
  47 {
  48     return (VirtIOBalloon *)vdev;
  49 }
  50
  51 static void balloon_page(void *addr, int deflate)
  52 {
  53 #if defined(__linux__)
  54     if (!kvm_enabled() || kvm_has_sync_mmu())
  55         madvise(addr, TARGET_PAGE_SIZE,
  56                 deflate ? MADV_WILLNEED : MADV_DONTNEED);
  57 #endif
  58 }
  59
  60 /*
  61  * reset_stats - Mark all items in the stats array as unset
  62  *
  63  * This function needs to be called at device intialization and before
  64  * before updating to a set of newly-generated stats.  This will ensure that no
  65  * stale values stick around in case the guest reports a subset of the supported
  66  * statistics.
  67  */
  68 static inline void reset_stats(VirtIOBalloon *dev)
  69 {
  70     int i;
  71     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
  72 }
  73
  74 static void stat_put(QDict *dict, const char *label, uint64_t val)
  75 {
  76     if (val != -1)
  77         qdict_put(dict, label, qint_from_int(val));
  78 }
  79
  80 static QObject *get_stats_qobject(VirtIOBalloon *dev)
  81 {
  82     QDict *dict = qdict_new();
  83     uint64_t actual = ram_size - ((uint64_t) dev->actual <<
  84                                   VIRTIO_BALLOON_PFN_SHIFT);
  85
  86     stat_put(dict, "actual", actual);
  87     stat_put(dict, "mem_swapped_in", dev->stats[VIRTIO_BALLOON_S_SWAP_IN]);
  88     stat_put(dict, "mem_swapped_out", dev->stats[VIRTIO_BALLOON_S_SWAP_OUT]);
  89     stat_put(dict, "major_page_faults", dev->stats[VIRTIO_BALLOON_S_MAJFLT]);
  90     stat_put(dict, "minor_page_faults", dev->stats[VIRTIO_BALLOON_S_MINFLT]);
  91     stat_put(dict, "free_mem", dev->stats[VIRTIO_BALLOON_S_MEMFREE]);
  92     stat_put(dict, "total_mem", dev->stats[VIRTIO_BALLOON_S_MEMTOT]);
  93
  94     return QOBJECT(dict);
  95 }
  96
  97 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
  98 {
  99     VirtIOBalloon *s = to_virtio_balloon(vdev);
 100     VirtQueueElement elem;
 101
 102     while (virtqueue_pop(vq, &elem)) {
 103         size_t offset = 0;
 104         uint32_t pfn;
 105
 106         while (iov_to_buf(elem.out_sg, elem.out_num, &pfn, offset, 4) == 4) {
 107             ram_addr_t pa;
 108             ram_addr_t addr;
 109
 110             pa = (ram_addr_t)ldl_p(&pfn) << VIRTIO_BALLOON_PFN_SHIFT;
 111             offset += 4;
 112
 113             addr = cpu_get_physical_page_desc(pa);
 114             if ((addr & ~TARGET_PAGE_MASK) != IO_MEM_RAM)
 115                 continue;
 116
 117             /* Using qemu_get_ram_ptr is bending the rules a bit, but
 118                should be OK because we only want a single page.  */
 119             balloon_page(qemu_get_ram_ptr(addr), !!(vq == s->dvq));
 120         }
 121
 122         virtqueue_push(vq, &elem, offset);
 123         virtio_notify(vdev, vq);
 124     }
 125 }
 126
 127 static void complete_stats_request(VirtIOBalloon *vb)
 128 {
 129     QObject *stats;
 130
 131     if (!vb->stats_opaque_callback_data)
 132         return;
 133
 134     stats = get_stats_qobject(vb);
 135     vb->stats_callback(vb->stats_opaque_callback_data, stats);
 136     qobject_decref(stats);
 137     vb->stats_opaque_callback_data = NULL;
 138     vb->stats_callback = NULL;
 139 }
 140
 141 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
 142 {
 143     VirtIOBalloon *s = DO_UPCAST(VirtIOBalloon, vdev, vdev);
 144     VirtQueueElement *elem = &s->stats_vq_elem;
 145     VirtIOBalloonStat stat;
 146     size_t offset = 0;
 147
 148     if (!virtqueue_pop(vq, elem)) {
 149         return;
 150     }
 151
 152     /* Initialize the stats to get rid of any stale values.  This is only
 153      * needed to handle the case where a guest supports fewer stats than it
 154      * used to (ie. it has booted into an old kernel).
 155      */
 156     reset_stats(s);
 157
 158     while (iov_to_buf(elem->out_sg, elem->out_num, &stat, offset, sizeof(stat))
 159            == sizeof(stat)) {
 160         uint16_t tag = tswap16(stat.tag);
 161         uint64_t val = tswap64(stat.val);
 162
 163         offset += sizeof(stat);
 164         if (tag < VIRTIO_BALLOON_S_NR)
 165             s->stats[tag] = val;
 166     }
 167     s->stats_vq_offset = offset;
 168
 169     complete_stats_request(s);
 170 }
 171
 172 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
 173 {
 174     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 175     struct virtio_balloon_config config;
 176
 177     config.num_pages = cpu_to_le32(dev->num_pages);
 178     config.actual = cpu_to_le32(dev->actual);
 179
 180     memcpy(config_data, &config, 8);
 181 }
 182
 183 static void virtio_balloon_set_config(VirtIODevice *vdev,
 184                                       const uint8_t *config_data)
 185 {
 186     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 187     struct virtio_balloon_config config;
 188     memcpy(&config, config_data, 8);
 189     dev->actual = config.actual;
 190 }
 191
 192 static uint32_t virtio_balloon_get_features(VirtIODevice *vdev, uint32_t f)
 193 {
 194     f |= (1 << VIRTIO_BALLOON_F_STATS_VQ);
 195     return f;
 196 }
 197
 198 static void virtio_balloon_to_target(void *opaque, ram_addr_t target,
 199                                      MonitorCompletion cb, void *cb_data)
 200 {
 201     VirtIOBalloon *dev = opaque;
 202
 203     if (target > ram_size)
 204         target = ram_size;
 205
 206     if (target) {
 207         dev->num_pages = (ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
 208         virtio_notify_config(&dev->vdev);
 209     } else {
 210         /* For now, only allow one request at a time.  This restriction can be
 211          * removed later by queueing callback and data pairs.
 212          */
 213         if (dev->stats_callback != NULL) {
 214             return;
 215         }
 216         dev->stats_callback = cb;
 217         dev->stats_opaque_callback_data = cb_data;
 218         if (dev->vdev.guest_features & (1 << VIRTIO_BALLOON_F_STATS_VQ)) {
 219             virtqueue_push(dev->svq, &dev->stats_vq_elem, dev->stats_vq_offset);
 220             virtio_notify(&dev->vdev, dev->svq);
 221         } else {
 222             /* Stats are not supported.  Clear out any stale values that might
 223              * have been set by a more featureful guest kernel.
 224              */
 225             reset_stats(dev);
 226             complete_stats_request(dev);
 227         }
 228     }
 229 }
 230
 231 static void virtio_balloon_save(QEMUFile *f, void *opaque)
 232 {
 233     VirtIOBalloon *s = opaque;
 234
 235     virtio_save(&s->vdev, f);
 236
 237     qemu_put_be32(f, s->num_pages);
 238     qemu_put_be32(f, s->actual);
 239 }
 240
 241 static int virtio_balloon_load(QEMUFile *f, void *opaque, int version_id)
 242 {
 243     VirtIOBalloon *s = opaque;
 244
 245     if (version_id != 1)
 246         return -EINVAL;
 247
 248     virtio_load(&s->vdev, f);
 249
 250     s->num_pages = qemu_get_be32(f);
 251     s->actual = qemu_get_be32(f);
 252     return 0;
 253 }
 254
 255 VirtIODevice *virtio_balloon_init(DeviceState *dev)
 256 {
 257     VirtIOBalloon *s;
 258
 259     s = (VirtIOBalloon *)virtio_common_init("virtio-balloon",
 260                                             VIRTIO_ID_BALLOON,
 261                                             8, sizeof(VirtIOBalloon));
 262
 263     s->vdev.get_config = virtio_balloon_get_config;
 264     s->vdev.set_config = virtio_balloon_set_config;
 265     s->vdev.get_features = virtio_balloon_get_features;
 266
 267     s->ivq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 268     s->dvq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 269     s->svq = virtio_add_queue(&s->vdev, 128, virtio_balloon_receive_stats);
 270
 271     reset_stats(s);
 272     qemu_add_balloon_handler(virtio_balloon_to_target, s);
 273
 274     register_savevm("virtio-balloon", -1, 1, virtio_balloon_save, virtio_balloon_load, s);
 275
 276     return &s->vdev;
 277 }