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