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 "qemu-common.h"
  15 #include "virtio.h"
  16 #include "pc.h"
  17 #include "sysemu.h"
  18 #include "cpu.h"
  19 #include "monitor.h"
  20 #include "balloon.h"
  21 #include "virtio-balloon.h"
  22 #include "kvm.h"
  23 #include "qlist.h"
  24 #include "qint.h"
  25 #include "qstring.h"
  26
  27 #if defined(__linux__)
  28 #include <sys/mman.h>
  29 #endif
  30
  31 typedef struct VirtIOBalloon
  32 {
  33     VirtIODevice vdev;
  34     VirtQueue *ivq, *dvq, *svq;
  35     uint32_t num_pages;
  36     uint32_t actual;
  37     uint64_t stats[VIRTIO_BALLOON_S_NR];
  38     VirtQueueElement stats_vq_elem;
  39     size_t stats_vq_offset;
  40     MonitorCompletion *stats_callback;
  41     void *stats_opaque_callback_data;
  42 } VirtIOBalloon;
  43
  44 static VirtIOBalloon *to_virtio_balloon(VirtIODevice *vdev)
  45 {
  46     return (VirtIOBalloon *)vdev;
  47 }
  48
  49 static void balloon_page(void *addr, int deflate)
  50 {
  51 #if defined(__linux__)
  52     if (!kvm_enabled() || kvm_has_sync_mmu())
  53         madvise(addr, TARGET_PAGE_SIZE,
  54                 deflate ? MADV_WILLNEED : MADV_DONTNEED);
  55 #endif
  56 }
  57
  58 /*
  59  * reset_stats - Mark all items in the stats array as unset
  60  *
  61  * This function needs to be called at device intialization and before
  62  * before updating to a set of newly-generated stats.  This will ensure that no
  63  * stale values stick around in case the guest reports a subset of the supported
  64  * statistics.
  65  */
  66 static inline void reset_stats(VirtIOBalloon *dev)
  67 {
  68     int i;
  69     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
  70 }
  71
  72 static void stat_put(QDict *dict, const char *label, uint64_t val)
  73 {
  74     if (val != -1)
  75         qdict_put(dict, label, qint_from_int(val));
  76 }
  77
  78 static QObject *get_stats_qobject(VirtIOBalloon *dev)
  79 {
  80     QDict *dict = qdict_new();
  81     uint32_t actual = ram_size - (dev->actual << VIRTIO_BALLOON_PFN_SHIFT);
  82
  83     stat_put(dict, "actual", actual);
  84     stat_put(dict, "mem_swapped_in", dev->stats[VIRTIO_BALLOON_S_SWAP_IN]);
  85     stat_put(dict, "mem_swapped_out", dev->stats[VIRTIO_BALLOON_S_SWAP_OUT]);
  86     stat_put(dict, "major_page_faults", dev->stats[VIRTIO_BALLOON_S_MAJFLT]);
  87     stat_put(dict, "minor_page_faults", dev->stats[VIRTIO_BALLOON_S_MINFLT]);
  88     stat_put(dict, "free_mem", dev->stats[VIRTIO_BALLOON_S_MEMFREE]);
  89     stat_put(dict, "total_mem", dev->stats[VIRTIO_BALLOON_S_MEMTOT]);
  90
  91     return QOBJECT(dict);
  92 }
  93
  94 /* FIXME: once we do a virtio refactoring, this will get subsumed into common
  95  * code */
  96 static size_t memcpy_from_iovector(void *data, size_t offset, size_t size,
  97                                    struct iovec *iov, int iovlen)
  98 {
  99     int i;
 100     uint8_t *ptr = data;
 101     size_t iov_off = 0;
 102     size_t data_off = 0;
 103
 104     for (i = 0; i < iovlen && size; i++) {
 105         if (offset < (iov_off + iov[i].iov_len)) {
 106             size_t len = MIN((iov_off + iov[i].iov_len) - offset , size);
 107
 108             memcpy(ptr + data_off, iov[i].iov_base + (offset - iov_off), len);
 109
 110             data_off += len;
 111             offset += len;
 112             size -= len;
 113         }
 114
 115         iov_off += iov[i].iov_len;
 116     }
 117
 118     return data_off;
 119 }
 120
 121 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 122 {
 123     VirtIOBalloon *s = to_virtio_balloon(vdev);
 124     VirtQueueElement elem;
 125
 126     while (virtqueue_pop(vq, &elem)) {
 127         size_t offset = 0;
 128         uint32_t pfn;
 129
 130         while (memcpy_from_iovector(&pfn, offset, 4,
 131                                     elem.out_sg, elem.out_num) == 4) {
 132             ram_addr_t pa;
 133             ram_addr_t addr;
 134
 135             pa = (ram_addr_t)ldl_p(&pfn) << VIRTIO_BALLOON_PFN_SHIFT;
 136             offset += 4;
 137
 138             addr = cpu_get_physical_page_desc(pa);
 139             if ((addr & ~TARGET_PAGE_MASK) != IO_MEM_RAM)
 140                 continue;
 141
 142             /* Using qemu_get_ram_ptr is bending the rules a bit, but
 143                should be OK because we only want a single page.  */
 144             balloon_page(qemu_get_ram_ptr(addr), !!(vq == s->dvq));
 145         }
 146
 147         virtqueue_push(vq, &elem, offset);
 148         virtio_notify(vdev, vq);
 149     }
 150 }
 151
 152 static void complete_stats_request(VirtIOBalloon *vb)
 153 {
 154     QObject *stats;
 155
 156     if (!vb->stats_opaque_callback_data)
 157         return;
 158
 159     stats = get_stats_qobject(vb);
 160     vb->stats_callback(vb->stats_opaque_callback_data, stats);
 161     qobject_decref(stats);
 162     vb->stats_opaque_callback_data = NULL;
 163     vb->stats_callback = NULL;
 164 }
 165
 166 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
 167 {
 168     VirtIOBalloon *s = DO_UPCAST(VirtIOBalloon, vdev, vdev);
 169     VirtQueueElement *elem = &s->stats_vq_elem;
 170     VirtIOBalloonStat stat;
 171     size_t offset = 0;
 172
 173     if (!virtqueue_pop(vq, elem)) {
 174         return;
 175     }
 176
 177     /* Initialize the stats to get rid of any stale values.  This is only
 178      * needed to handle the case where a guest supports fewer stats than it
 179      * used to (ie. it has booted into an old kernel).
 180      */
 181     reset_stats(s);
 182
 183     while (memcpy_from_iovector(&stat, offset, sizeof(stat), elem->out_sg,
 184                                 elem->out_num) == sizeof(stat)) {
 185         uint16_t tag = tswap16(stat.tag);
 186         uint64_t val = tswap64(stat.val);
 187
 188         offset += sizeof(stat);
 189         if (tag < VIRTIO_BALLOON_S_NR)
 190             s->stats[tag] = val;
 191     }
 192     s->stats_vq_offset = offset;
 193
 194     complete_stats_request(s);
 195 }
 196
 197 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
 198 {
 199     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 200     struct virtio_balloon_config config;
 201
 202     config.num_pages = cpu_to_le32(dev->num_pages);
 203     config.actual = cpu_to_le32(dev->actual);
 204
 205     memcpy(config_data, &config, 8);
 206 }
 207
 208 static void virtio_balloon_set_config(VirtIODevice *vdev,
 209                                       const uint8_t *config_data)
 210 {
 211     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 212     struct virtio_balloon_config config;
 213     memcpy(&config, config_data, 8);
 214     dev->actual = config.actual;
 215 }
 216
 217 static uint32_t virtio_balloon_get_features(VirtIODevice *vdev, uint32_t f)
 218 {
 219     f |= (1 << VIRTIO_BALLOON_F_STATS_VQ);
 220     return f;
 221 }
 222
 223 static void virtio_balloon_to_target(void *opaque, ram_addr_t target,
 224                                      MonitorCompletion cb, void *cb_data)
 225 {
 226     VirtIOBalloon *dev = opaque;
 227
 228     if (target > ram_size)
 229         target = ram_size;
 230
 231     if (target) {
 232         dev->num_pages = (ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
 233         virtio_notify_config(&dev->vdev);
 234     } else {
 235         /* For now, only allow one request at a time.  This restriction can be
 236          * removed later by queueing callback and data pairs.
 237          */
 238         if (dev->stats_callback != NULL) {
 239             return;
 240         }
 241         dev->stats_callback = cb;
 242         dev->stats_opaque_callback_data = cb_data;
 243         if (dev->vdev.guest_features & (1 << VIRTIO_BALLOON_F_STATS_VQ)) {
 244             virtqueue_push(dev->svq, &dev->stats_vq_elem, dev->stats_vq_offset);
 245             virtio_notify(&dev->vdev, dev->svq);
 246         } else {
 247             /* Stats are not supported.  Clear out any stale values that might
 248              * have been set by a more featureful guest kernel.
 249              */
 250             reset_stats(dev);
 251             complete_stats_request(dev);
 252         }
 253     }
 254 }
 255
 256 static void virtio_balloon_save(QEMUFile *f, void *opaque)
 257 {
 258     VirtIOBalloon *s = opaque;
 259
 260     virtio_save(&s->vdev, f);
 261
 262     qemu_put_be32(f, s->num_pages);
 263     qemu_put_be32(f, s->actual);
 264     qemu_put_buffer(f, (uint8_t *)&s->stats_vq_elem, sizeof(VirtQueueElement));
 265     qemu_put_buffer(f, (uint8_t *)&s->stats_vq_offset, sizeof(size_t));
 266     qemu_put_buffer(f, (uint8_t *)&s->stats_callback, sizeof(MonitorCompletion));
 267     qemu_put_buffer(f, (uint8_t *)&s->stats_opaque_callback_data, sizeof(void));
 268 }
 269
 270 static int virtio_balloon_load(QEMUFile *f, void *opaque, int version_id)
 271 {
 272     VirtIOBalloon *s = opaque;
 273
 274     if (version_id != 1)
 275         return -EINVAL;
 276
 277     virtio_load(&s->vdev, f);
 278
 279     s->num_pages = qemu_get_be32(f);
 280     s->actual = qemu_get_be32(f);
 281     qemu_get_buffer(f, (uint8_t *)&s->stats_vq_elem, sizeof(VirtQueueElement));
 282     qemu_get_buffer(f, (uint8_t *)&s->stats_vq_offset, sizeof(size_t));
 283     qemu_get_buffer(f, (uint8_t *)&s->stats_callback, sizeof(MonitorCompletion));
 284     qemu_get_buffer(f, (uint8_t *)&s->stats_opaque_callback_data, sizeof(void));
 285
 286     return 0;
 287 }
 288
 289 VirtIODevice *virtio_balloon_init(DeviceState *dev)
 290 {
 291     VirtIOBalloon *s;
 292
 293     s = (VirtIOBalloon *)virtio_common_init("virtio-balloon",
 294                                             VIRTIO_ID_BALLOON,
 295                                             8, sizeof(VirtIOBalloon));
 296
 297     s->vdev.get_config = virtio_balloon_get_config;
 298     s->vdev.set_config = virtio_balloon_set_config;
 299     s->vdev.get_features = virtio_balloon_get_features;
 300
 301     s->ivq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 302     s->dvq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 303     s->svq = virtio_add_queue(&s->vdev, 128, virtio_balloon_receive_stats);
 304
 305     reset_stats(s);
 306     qemu_add_balloon_handler(virtio_balloon_to_target, s);
 307
 308     register_savevm("virtio-balloon", -1, 1, virtio_balloon_save, virtio_balloon_load, s);
 309
 310     return &s->vdev;
 311 }