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 /* Disable guest-provided stats by now (https://bugzilla.redhat.com/show_bug.cgi?id=623903) */
  34 #define ENABLE_GUEST_STATS   0
  35
  36
  37 typedef struct VirtIOBalloon
  38 {
  39     VirtIODevice vdev;
  40     VirtQueue *ivq, *dvq, *svq;
  41     uint32_t num_pages;
  42     uint32_t actual;
  43     uint64_t stats[VIRTIO_BALLOON_S_NR];
  44     VirtQueueElement stats_vq_elem;
  45     size_t stats_vq_offset;
  46     MonitorCompletion *stats_callback;
  47     void *stats_opaque_callback_data;
  48 } VirtIOBalloon;
  49
  50 static VirtIOBalloon *to_virtio_balloon(VirtIODevice *vdev)
  51 {
  52     return (VirtIOBalloon *)vdev;
  53 }
  54
  55 static void balloon_page(void *addr, int deflate)
  56 {
  57 #if defined(__linux__)
  58     if (!kvm_enabled() || kvm_has_sync_mmu())
  59         qemu_madvise(addr, TARGET_PAGE_SIZE,
  60                 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
  61 #endif
  62 }
  63
  64 /*
  65  * reset_stats - Mark all items in the stats array as unset
  66  *
  67  * This function needs to be called at device intialization and before
  68  * before updating to a set of newly-generated stats.  This will ensure that no
  69  * stale values stick around in case the guest reports a subset of the supported
  70  * statistics.
  71  */
  72 static inline void reset_stats(VirtIOBalloon *dev)
  73 {
  74     int i;
  75     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
  76 }
  77
  78 static void stat_put(QDict *dict, const char *label, uint64_t val)
  79 {
  80     if (val != -1)
  81         qdict_put(dict, label, qint_from_int(val));
  82 }
  83
  84 static QObject *get_stats_qobject(VirtIOBalloon *dev)
  85 {
  86     QDict *dict = qdict_new();
  87     uint64_t actual = ram_size - ((uint64_t) dev->actual <<
  88                                   VIRTIO_BALLOON_PFN_SHIFT);
  89
  90     stat_put(dict, "actual", actual);
  91 #if ENABLE_GUEST_STATS
  92     stat_put(dict, "mem_swapped_in", dev->stats[VIRTIO_BALLOON_S_SWAP_IN]);
  93     stat_put(dict, "mem_swapped_out", dev->stats[VIRTIO_BALLOON_S_SWAP_OUT]);
  94     stat_put(dict, "major_page_faults", dev->stats[VIRTIO_BALLOON_S_MAJFLT]);
  95     stat_put(dict, "minor_page_faults", dev->stats[VIRTIO_BALLOON_S_MINFLT]);
  96     stat_put(dict, "free_mem", dev->stats[VIRTIO_BALLOON_S_MEMFREE]);
  97     stat_put(dict, "total_mem", dev->stats[VIRTIO_BALLOON_S_MEMTOT]);
  98 #endif
  99
 100     return QOBJECT(dict);
 101 }
 102
 103 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 104 {
 105     VirtIOBalloon *s = to_virtio_balloon(vdev);
 106     VirtQueueElement elem;
 107
 108     while (virtqueue_pop(vq, &elem)) {
 109         size_t offset = 0;
 110         uint32_t pfn;
 111
 112         while (iov_to_buf(elem.out_sg, elem.out_num, &pfn, offset, 4) == 4) {
 113             ram_addr_t pa;
 114             ram_addr_t addr;
 115
 116             pa = (ram_addr_t)ldl_p(&pfn) << VIRTIO_BALLOON_PFN_SHIFT;
 117             offset += 4;
 118
 119             addr = cpu_get_physical_page_desc(pa);
 120             if ((addr & ~TARGET_PAGE_MASK) != IO_MEM_RAM)
 121                 continue;
 122
 123             /* Using qemu_get_ram_ptr is bending the rules a bit, but
 124                should be OK because we only want a single page.  */
 125             balloon_page(qemu_get_ram_ptr(addr), !!(vq == s->dvq));
 126         }
 127
 128         virtqueue_push(vq, &elem, offset);
 129         virtio_notify(vdev, vq);
 130     }
 131 }
 132
 133 static void complete_stats_request(VirtIOBalloon *vb)
 134 {
 135     QObject *stats;
 136
 137     if (!vb->stats_opaque_callback_data)
 138         return;
 139
 140     stats = get_stats_qobject(vb);
 141     vb->stats_callback(vb->stats_opaque_callback_data, stats);
 142     qobject_decref(stats);
 143     vb->stats_opaque_callback_data = NULL;
 144     vb->stats_callback = NULL;
 145 }
 146
 147 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
 148 {
 149     VirtIOBalloon *s = DO_UPCAST(VirtIOBalloon, vdev, vdev);
 150     VirtQueueElement *elem = &s->stats_vq_elem;
 151     VirtIOBalloonStat stat;
 152     size_t offset = 0;
 153
 154     if (!virtqueue_pop(vq, elem)) {
 155         return;
 156     }
 157
 158     /* Initialize the stats to get rid of any stale values.  This is only
 159      * needed to handle the case where a guest supports fewer stats than it
 160      * used to (ie. it has booted into an old kernel).
 161      */
 162     reset_stats(s);
 163
 164     while (iov_to_buf(elem->out_sg, elem->out_num, &stat, offset, sizeof(stat))
 165            == sizeof(stat)) {
 166         uint16_t tag = tswap16(stat.tag);
 167         uint64_t val = tswap64(stat.val);
 168
 169         offset += sizeof(stat);
 170         if (tag < VIRTIO_BALLOON_S_NR)
 171             s->stats[tag] = val;
 172     }
 173     s->stats_vq_offset = offset;
 174
 175     complete_stats_request(s);
 176 }
 177
 178 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
 179 {
 180     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 181     struct virtio_balloon_config config;
 182
 183     config.num_pages = cpu_to_le32(dev->num_pages);
 184     config.actual = cpu_to_le32(dev->actual);
 185
 186     memcpy(config_data, &config, 8);
 187 }
 188
 189 static void virtio_balloon_set_config(VirtIODevice *vdev,
 190                                       const uint8_t *config_data)
 191 {
 192     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 193     struct virtio_balloon_config config;
 194     memcpy(&config, config_data, 8);
 195     dev->actual = config.actual;
 196 }
 197
 198 static uint32_t virtio_balloon_get_features(VirtIODevice *vdev, uint32_t f)
 199 {
 200     f |= (1 << VIRTIO_BALLOON_F_STATS_VQ);
 201     return f;
 202 }
 203
 204 static void virtio_balloon_to_target(void *opaque, ram_addr_t target,
 205                                      MonitorCompletion cb, void *cb_data)
 206 {
 207     VirtIOBalloon *dev = opaque;
 208
 209     if (target > ram_size)
 210         target = ram_size;
 211
 212     if (target) {
 213         dev->num_pages = (ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
 214         virtio_notify_config(&dev->vdev);
 215     } else {
 216         /* For now, only allow one request at a time.  This restriction can be
 217          * removed later by queueing callback and data pairs.
 218          */
 219         if (dev->stats_callback != NULL) {
 220             return;
 221         }
 222         dev->stats_callback = cb;
 223         dev->stats_opaque_callback_data = cb_data;
 224         if (ENABLE_GUEST_STATS && (dev->vdev.guest_features & (1 << VIRTIO_BALLOON_F_STATS_VQ))) {
 225             virtqueue_push(dev->svq, &dev->stats_vq_elem, dev->stats_vq_offset);
 226             virtio_notify(&dev->vdev, dev->svq);
 227         } else {
 228             /* Stats are not supported.  Clear out any stale values that might
 229              * have been set by a more featureful guest kernel.
 230              */
 231             reset_stats(dev);
 232             complete_stats_request(dev);
 233         }
 234     }
 235 }
 236
 237 static void virtio_balloon_save(QEMUFile *f, void *opaque)
 238 {
 239     VirtIOBalloon *s = opaque;
 240
 241     virtio_save(&s->vdev, f);
 242
 243     qemu_put_be32(f, s->num_pages);
 244     qemu_put_be32(f, s->actual);
 245 }
 246
 247 static int virtio_balloon_load(QEMUFile *f, void *opaque, int version_id)
 248 {
 249     VirtIOBalloon *s = opaque;
 250
 251     if (version_id != 1)
 252         return -EINVAL;
 253
 254     virtio_load(&s->vdev, f);
 255
 256     s->num_pages = qemu_get_be32(f);
 257     s->actual = qemu_get_be32(f);
 258     return 0;
 259 }
 260
 261 VirtIODevice *virtio_balloon_init(DeviceState *dev)
 262 {
 263     VirtIOBalloon *s;
 264
 265     s = (VirtIOBalloon *)virtio_common_init("virtio-balloon",
 266                                             VIRTIO_ID_BALLOON,
 267                                             8, sizeof(VirtIOBalloon));
 268
 269     s->vdev.get_config = virtio_balloon_get_config;
 270     s->vdev.set_config = virtio_balloon_set_config;
 271     s->vdev.get_features = virtio_balloon_get_features;
 272
 273     s->ivq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 274     s->dvq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 275     s->svq = virtio_add_queue(&s->vdev, 128, virtio_balloon_receive_stats);
 276
 277     reset_stats(s);
 278     qemu_add_balloon_handler(virtio_balloon_to_target, s);
 279
 280     register_savevm(dev, "virtio-balloon", -1, 1,
 281                     virtio_balloon_save, virtio_balloon_load, s);
 282
 283     return &s->vdev;
 284 }