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 "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 /* Disable guest-provided stats by now (https://bugzilla.redhat.com/show_bug.cgi?id=623903) */
  32 #define ENABLE_GUEST_STATS   0
  33
  34
  35 typedef struct VirtIOBalloon
  36 {
  37     VirtIODevice vdev;
  38     VirtQueue *ivq, *dvq, *svq;
  39     uint32_t num_pages;
  40     uint32_t actual;
  41     uint64_t stats[VIRTIO_BALLOON_S_NR];
  42     VirtQueueElement stats_vq_elem;
  43     size_t stats_vq_offset;
  44     MonitorCompletion *stats_callback;
  45     void *stats_opaque_callback_data;
  46 } VirtIOBalloon;
  47
  48 static VirtIOBalloon *to_virtio_balloon(VirtIODevice *vdev)
  49 {
  50     return (VirtIOBalloon *)vdev;
  51 }
  52
  53 static void balloon_page(void *addr, int deflate)
  54 {
  55 #if defined(__linux__)
  56     if (!kvm_enabled() || kvm_has_sync_mmu())
  57         qemu_madvise(addr, TARGET_PAGE_SIZE,
  58                 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
  59 #endif
  60 }
  61
  62 /*
  63  * reset_stats - Mark all items in the stats array as unset
  64  *
  65  * This function needs to be called at device intialization and before
  66  * before updating to a set of newly-generated stats.  This will ensure that no
  67  * stale values stick around in case the guest reports a subset of the supported
  68  * statistics.
  69  */
  70 static inline void reset_stats(VirtIOBalloon *dev)
  71 {
  72     int i;
  73     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
  74 }
  75
  76 static void stat_put(QDict *dict, const char *label, uint64_t val)
  77 {
  78     if (val != -1)
  79         qdict_put(dict, label, qint_from_int(val));
  80 }
  81
  82 static QObject *get_stats_qobject(VirtIOBalloon *dev)
  83 {
  84     QDict *dict = qdict_new();
  85     uint64_t actual = ram_size - ((uint64_t) dev->actual <<
  86                                   VIRTIO_BALLOON_PFN_SHIFT);
  87
  88     stat_put(dict, "actual", actual);
  89 #if ENABLE_GUEST_STATS
  90     stat_put(dict, "mem_swapped_in", dev->stats[VIRTIO_BALLOON_S_SWAP_IN]);
  91     stat_put(dict, "mem_swapped_out", dev->stats[VIRTIO_BALLOON_S_SWAP_OUT]);
  92     stat_put(dict, "major_page_faults", dev->stats[VIRTIO_BALLOON_S_MAJFLT]);
  93     stat_put(dict, "minor_page_faults", dev->stats[VIRTIO_BALLOON_S_MINFLT]);
  94     stat_put(dict, "free_mem", dev->stats[VIRTIO_BALLOON_S_MEMFREE]);
  95     stat_put(dict, "total_mem", dev->stats[VIRTIO_BALLOON_S_MEMTOT]);
  96 #endif
  97
  98     return QOBJECT(dict);
  99 }
 100
 101 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 102 {
 103     VirtIOBalloon *s = to_virtio_balloon(vdev);
 104     VirtQueueElement elem;
 105
 106     while (virtqueue_pop(vq, &elem)) {
 107         size_t offset = 0;
 108         uint32_t pfn;
 109
 110         while (iov_to_buf(elem.out_sg, elem.out_num, &pfn, offset, 4) == 4) {
 111             ram_addr_t pa;
 112             ram_addr_t addr;
 113
 114             pa = (ram_addr_t)ldl_p(&pfn) << VIRTIO_BALLOON_PFN_SHIFT;
 115             offset += 4;
 116
 117             addr = cpu_get_physical_page_desc(pa);
 118             if ((addr & ~TARGET_PAGE_MASK) != IO_MEM_RAM)
 119                 continue;
 120
 121             /* Using qemu_get_ram_ptr is bending the rules a bit, but
 122                should be OK because we only want a single page.  */
 123             balloon_page(qemu_get_ram_ptr(addr), !!(vq == s->dvq));
 124         }
 125
 126         virtqueue_push(vq, &elem, offset);
 127         virtio_notify(vdev, vq);
 128     }
 129 }
 130
 131 static void complete_stats_request(VirtIOBalloon *vb)
 132 {
 133     QObject *stats;
 134
 135     if (!vb->stats_opaque_callback_data)
 136         return;
 137
 138     stats = get_stats_qobject(vb);
 139     vb->stats_callback(vb->stats_opaque_callback_data, stats);
 140     qobject_decref(stats);
 141     vb->stats_opaque_callback_data = NULL;
 142     vb->stats_callback = NULL;
 143 }
 144
 145 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
 146 {
 147     VirtIOBalloon *s = DO_UPCAST(VirtIOBalloon, vdev, vdev);
 148     VirtQueueElement *elem = &s->stats_vq_elem;
 149     VirtIOBalloonStat stat;
 150     size_t offset = 0;
 151
 152     if (!virtqueue_pop(vq, elem)) {
 153         return;
 154     }
 155
 156     /* Initialize the stats to get rid of any stale values.  This is only
 157      * needed to handle the case where a guest supports fewer stats than it
 158      * used to (ie. it has booted into an old kernel).
 159      */
 160     reset_stats(s);
 161
 162     while (iov_to_buf(elem->out_sg, elem->out_num, &stat, offset, sizeof(stat))
 163            == sizeof(stat)) {
 164         uint16_t tag = tswap16(stat.tag);
 165         uint64_t val = tswap64(stat.val);
 166
 167         offset += sizeof(stat);
 168         if (tag < VIRTIO_BALLOON_S_NR)
 169             s->stats[tag] = val;
 170     }
 171     s->stats_vq_offset = offset;
 172
 173     complete_stats_request(s);
 174 }
 175
 176 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
 177 {
 178     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 179     struct virtio_balloon_config config;
 180
 181     config.num_pages = cpu_to_le32(dev->num_pages);
 182     config.actual = cpu_to_le32(dev->actual);
 183
 184     memcpy(config_data, &config, 8);
 185 }
 186
 187 static void virtio_balloon_set_config(VirtIODevice *vdev,
 188                                       const uint8_t *config_data)
 189 {
 190     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 191     struct virtio_balloon_config config;
 192     memcpy(&config, config_data, 8);
 193     dev->actual = le32_to_cpu(config.actual);
 194 }
 195
 196 static uint32_t virtio_balloon_get_features(VirtIODevice *vdev, uint32_t f)
 197 {
 198     f |= (1 << VIRTIO_BALLOON_F_STATS_VQ);
 199     return f;
 200 }
 201
 202 static void virtio_balloon_to_target(void *opaque, ram_addr_t target,
 203                                      MonitorCompletion cb, void *cb_data)
 204 {
 205     VirtIOBalloon *dev = opaque;
 206
 207     if (target > ram_size)
 208         target = ram_size;
 209
 210     if (target) {
 211         dev->num_pages = (ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
 212         virtio_notify_config(&dev->vdev);
 213     } else {
 214         /* For now, only allow one request at a time.  This restriction can be
 215          * removed later by queueing callback and data pairs.
 216          */
 217         if (dev->stats_callback != NULL) {
 218             return;
 219         }
 220         dev->stats_callback = cb;
 221         dev->stats_opaque_callback_data = cb_data;
 222         if (ENABLE_GUEST_STATS && (dev->vdev.guest_features & (1 << VIRTIO_BALLOON_F_STATS_VQ))) {
 223             virtqueue_push(dev->svq, &dev->stats_vq_elem, dev->stats_vq_offset);
 224             virtio_notify(&dev->vdev, dev->svq);
 225         } else {
 226             /* Stats are not supported.  Clear out any stale values that might
 227              * have been set by a more featureful guest kernel.
 228              */
 229             reset_stats(dev);
 230             complete_stats_request(dev);
 231         }
 232     }
 233 }
 234
 235 static void virtio_balloon_save(QEMUFile *f, void *opaque)
 236 {
 237     VirtIOBalloon *s = opaque;
 238
 239     virtio_save(&s->vdev, f);
 240
 241     qemu_put_be32(f, s->num_pages);
 242     qemu_put_be32(f, s->actual);
 243 }
 244
 245 static int virtio_balloon_load(QEMUFile *f, void *opaque, int version_id)
 246 {
 247     VirtIOBalloon *s = opaque;
 248
 249     if (version_id != 1)
 250         return -EINVAL;
 251
 252     virtio_load(&s->vdev, f);
 253
 254     s->num_pages = qemu_get_be32(f);
 255     s->actual = qemu_get_be32(f);
 256     return 0;
 257 }
 258
 259 VirtIODevice *virtio_balloon_init(DeviceState *dev)
 260 {
 261     VirtIOBalloon *s;
 262
 263     s = (VirtIOBalloon *)virtio_common_init("virtio-balloon",
 264                                             VIRTIO_ID_BALLOON,
 265                                             8, sizeof(VirtIOBalloon));
 266
 267     s->vdev.get_config = virtio_balloon_get_config;
 268     s->vdev.set_config = virtio_balloon_set_config;
 269     s->vdev.get_features = virtio_balloon_get_features;
 270
 271     s->ivq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 272     s->dvq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 273     s->svq = virtio_add_queue(&s->vdev, 128, virtio_balloon_receive_stats);
 274
 275     reset_stats(s);
 276     qemu_add_balloon_handler(virtio_balloon_to_target, s);
 277
 278     register_savevm(dev, "virtio-balloon", -1, 1,
 279                     virtio_balloon_save, virtio_balloon_load, s);
 280
 281     return &s->vdev;
 282 }