hw/virtio-balloon.c

   1 /*
   2  * Virtio Balloon Device
   3  *
   4  * Copyright IBM, Corp. 2008
   5  * Copyright (C) 2011 Red Hat, Inc.
   6  * Copyright (C) 2011 Amit Shah <amit.shah@redhat.com>
   7  *
   8  * Authors:
   9  *  Anthony Liguori   <aliguori@us.ibm.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.  See
  12  * the COPYING file in the top-level directory.
  13  *
  14  */
  15
  16 #include "iov.h"
  17 #include "qemu-common.h"
  18 #include "virtio.h"
  19 #include "pc.h"
  20 #include "cpu.h"
  21 #include "monitor.h"
  22 #include "balloon.h"
  23 #include "virtio-balloon.h"
  24 #include "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     DeviceState *qdev;
  49 } VirtIOBalloon;
  50
  51 static VirtIOBalloon *to_virtio_balloon(VirtIODevice *vdev)
  52 {
  53     return (VirtIOBalloon *)vdev;
  54 }
  55
  56 static void balloon_page(void *addr, int deflate)
  57 {
  58 #if defined(__linux__)
  59     if (!kvm_enabled() || kvm_has_sync_mmu())
  60         qemu_madvise(addr, TARGET_PAGE_SIZE,
  61                 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
  62 #endif
  63 }
  64
  65 /*
  66  * reset_stats - Mark all items in the stats array as unset
  67  *
  68  * This function needs to be called at device intialization and before
  69  * before updating to a set of newly-generated stats.  This will ensure that no
  70  * stale values stick around in case the guest reports a subset of the supported
  71  * statistics.
  72  */
  73 static inline void reset_stats(VirtIOBalloon *dev)
  74 {
  75     int i;
  76     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
  77 }
  78
  79 static void stat_put(QDict *dict, const char *label, uint64_t val)
  80 {
  81     if (val != -1)
  82         qdict_put(dict, label, qint_from_int(val));
  83 }
  84
  85 static QObject *get_stats_qobject(VirtIOBalloon *dev)
  86 {
  87     QDict *dict = qdict_new();
  88     uint64_t actual = ram_size - ((uint64_t) dev->actual <<
  89                                   VIRTIO_BALLOON_PFN_SHIFT);
  90
  91     stat_put(dict, "actual", actual);
  92 #if ENABLE_GUEST_STATS
  93     stat_put(dict, "mem_swapped_in", dev->stats[VIRTIO_BALLOON_S_SWAP_IN]);
  94     stat_put(dict, "mem_swapped_out", dev->stats[VIRTIO_BALLOON_S_SWAP_OUT]);
  95     stat_put(dict, "major_page_faults", dev->stats[VIRTIO_BALLOON_S_MAJFLT]);
  96     stat_put(dict, "minor_page_faults", dev->stats[VIRTIO_BALLOON_S_MINFLT]);
  97     stat_put(dict, "free_mem", dev->stats[VIRTIO_BALLOON_S_MEMFREE]);
  98     stat_put(dict, "total_mem", dev->stats[VIRTIO_BALLOON_S_MEMTOT]);
  99 #endif
 100
 101     return QOBJECT(dict);
 102 }
 103
 104 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 105 {
 106     VirtIOBalloon *s = to_virtio_balloon(vdev);
 107     VirtQueueElement elem;
 108
 109     while (virtqueue_pop(vq, &elem)) {
 110         size_t offset = 0;
 111         uint32_t pfn;
 112
 113         while (iov_to_buf(elem.out_sg, elem.out_num, &pfn, offset, 4) == 4) {
 114             ram_addr_t pa;
 115             ram_addr_t addr;
 116
 117             pa = (ram_addr_t)ldl_p(&pfn) << VIRTIO_BALLOON_PFN_SHIFT;
 118             offset += 4;
 119
 120             addr = cpu_get_physical_page_desc(pa);
 121             if ((addr & ~TARGET_PAGE_MASK) != IO_MEM_RAM)
 122                 continue;
 123
 124             /* Using qemu_get_ram_ptr is bending the rules a bit, but
 125                should be OK because we only want a single page.  */
 126             balloon_page(qemu_get_ram_ptr(addr), !!(vq == s->dvq));
 127         }
 128
 129         virtqueue_push(vq, &elem, offset);
 130         virtio_notify(vdev, vq);
 131     }
 132 }
 133
 134 static void complete_stats_request(VirtIOBalloon *vb)
 135 {
 136     QObject *stats;
 137
 138     if (!vb->stats_opaque_callback_data)
 139         return;
 140
 141     stats = get_stats_qobject(vb);
 142     vb->stats_callback(vb->stats_opaque_callback_data, stats);
 143     qobject_decref(stats);
 144     vb->stats_opaque_callback_data = NULL;
 145     vb->stats_callback = NULL;
 146 }
 147
 148 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
 149 {
 150     VirtIOBalloon *s = DO_UPCAST(VirtIOBalloon, vdev, vdev);
 151     VirtQueueElement *elem = &s->stats_vq_elem;
 152     VirtIOBalloonStat stat;
 153     size_t offset = 0;
 154
 155     if (!virtqueue_pop(vq, elem)) {
 156         return;
 157     }
 158
 159     /* Initialize the stats to get rid of any stale values.  This is only
 160      * needed to handle the case where a guest supports fewer stats than it
 161      * used to (ie. it has booted into an old kernel).
 162      */
 163     reset_stats(s);
 164
 165     while (iov_to_buf(elem->out_sg, elem->out_num, &stat, offset, sizeof(stat))
 166            == sizeof(stat)) {
 167         uint16_t tag = tswap16(stat.tag);
 168         uint64_t val = tswap64(stat.val);
 169
 170         offset += sizeof(stat);
 171         if (tag < VIRTIO_BALLOON_S_NR)
 172             s->stats[tag] = val;
 173     }
 174     s->stats_vq_offset = offset;
 175
 176     complete_stats_request(s);
 177 }
 178
 179 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
 180 {
 181     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 182     struct virtio_balloon_config config;
 183
 184     config.num_pages = cpu_to_le32(dev->num_pages);
 185     config.actual = cpu_to_le32(dev->actual);
 186
 187     memcpy(config_data, &config, 8);
 188 }
 189
 190 static void virtio_balloon_set_config(VirtIODevice *vdev,
 191                                       const uint8_t *config_data)
 192 {
 193     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 194     struct virtio_balloon_config config;
 195     memcpy(&config, config_data, 8);
 196     dev->actual = le32_to_cpu(config.actual);
 197 }
 198
 199 static uint32_t virtio_balloon_get_features(VirtIODevice *vdev, uint32_t f)
 200 {
 201     f |= (1 << VIRTIO_BALLOON_F_STATS_VQ);
 202     return f;
 203 }
 204
 205 static void virtio_balloon_stat(void *opaque, MonitorCompletion cb,
 206                                 void *cb_data)
 207 {
 208     VirtIOBalloon *dev = opaque;
 209
 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
 219     if (ENABLE_GUEST_STATS
 220         && (dev->vdev.guest_features & (1 << VIRTIO_BALLOON_F_STATS_VQ))) {
 221         virtqueue_push(dev->svq, &dev->stats_vq_elem, dev->stats_vq_offset);
 222         virtio_notify(&dev->vdev, dev->svq);
 223         return;
 224     }
 225
 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 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
 234 {
 235     VirtIOBalloon *dev = opaque;
 236
 237     if (target > ram_size) {
 238         target = ram_size;
 239     }
 240     if (target) {
 241         dev->num_pages = (ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
 242         virtio_notify_config(&dev->vdev);
 243     }
 244 }
 245
 246 static void virtio_balloon_save(QEMUFile *f, void *opaque)
 247 {
 248     VirtIOBalloon *s = opaque;
 249
 250     virtio_save(&s->vdev, f);
 251
 252     qemu_put_be32(f, s->num_pages);
 253     qemu_put_be32(f, s->actual);
 254 }
 255
 256 static int virtio_balloon_load(QEMUFile *f, void *opaque, int version_id)
 257 {
 258     VirtIOBalloon *s = opaque;
 259
 260     if (version_id != 1)
 261         return -EINVAL;
 262
 263     virtio_load(&s->vdev, f);
 264
 265     s->num_pages = qemu_get_be32(f);
 266     s->actual = qemu_get_be32(f);
 267     return 0;
 268 }
 269
 270 VirtIODevice *virtio_balloon_init(DeviceState *dev)
 271 {
 272     VirtIOBalloon *s;
 273     int ret;
 274
 275     s = (VirtIOBalloon *)virtio_common_init("virtio-balloon",
 276                                             VIRTIO_ID_BALLOON,
 277                                             8, sizeof(VirtIOBalloon));
 278
 279     s->vdev.get_config = virtio_balloon_get_config;
 280     s->vdev.set_config = virtio_balloon_set_config;
 281     s->vdev.get_features = virtio_balloon_get_features;
 282
 283     ret = qemu_add_balloon_handler(virtio_balloon_to_target,
 284                                    virtio_balloon_stat, s);
 285     if (ret < 0) {
 286         virtio_cleanup(&s->vdev);
 287         return NULL;
 288     }
 289
 290     s->ivq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 291     s->dvq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 292     s->svq = virtio_add_queue(&s->vdev, 128, virtio_balloon_receive_stats);
 293
 294     reset_stats(s);
 295
 296     s->qdev = dev;
 297     register_savevm(dev, "virtio-balloon", -1, 1,
 298                     virtio_balloon_save, virtio_balloon_load, s);
 299
 300     return &s->vdev;
 301 }
 302
 303 void virtio_balloon_exit(VirtIODevice *vdev)
 304 {
 305     VirtIOBalloon *s = DO_UPCAST(VirtIOBalloon, vdev, vdev);
 306
 307     qemu_remove_balloon_handler(s);
 308     unregister_savevm(s->qdev, "virtio-balloon", s);
 309     virtio_cleanup(vdev);
 310 }