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 /* Disable guest-provided stats by now (https://bugzilla.redhat.com/show_bug.cgi?id=623903) */
  33 #define ENABLE_GUEST_STATS   0
  34
  35
  36 typedef struct VirtIOBalloon
  37 {
  38     VirtIODevice vdev;
  39     VirtQueue *ivq, *dvq, *svq;
  40     uint32_t num_pages;
  41     uint32_t actual;
  42     uint64_t stats[VIRTIO_BALLOON_S_NR];
  43     VirtQueueElement stats_vq_elem;
  44     size_t stats_vq_offset;
  45     MonitorCompletion *stats_callback;
  46     void *stats_opaque_callback_data;
  47 } VirtIOBalloon;
  48
  49 static VirtIOBalloon *to_virtio_balloon(VirtIODevice *vdev)
  50 {
  51     return (VirtIOBalloon *)vdev;
  52 }
  53
  54 static void balloon_page(void *addr, int deflate)
  55 {
  56 #if defined(__linux__)
  57     if (!kvm_enabled() || kvm_has_sync_mmu())
  58         qemu_madvise(addr, TARGET_PAGE_SIZE,
  59                 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
  60 #endif
  61 }
  62
  63 /*
  64  * reset_stats - Mark all items in the stats array as unset
  65  *
  66  * This function needs to be called at device intialization and before
  67  * before updating to a set of newly-generated stats.  This will ensure that no
  68  * stale values stick around in case the guest reports a subset of the supported
  69  * statistics.
  70  */
  71 static inline void reset_stats(VirtIOBalloon *dev)
  72 {
  73     int i;
  74     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
  75 }
  76
  77 static void stat_put(QDict *dict, const char *label, uint64_t val)
  78 {
  79     if (val != -1)
  80         qdict_put(dict, label, qint_from_int(val));
  81 }
  82
  83 static QObject *get_stats_qobject(VirtIOBalloon *dev)
  84 {
  85     QDict *dict = qdict_new();
  86     uint64_t actual = ram_size - ((uint64_t) dev->actual <<
  87                                   VIRTIO_BALLOON_PFN_SHIFT);
  88
  89     stat_put(dict, "actual", actual);
  90 #if ENABLE_GUEST_STATS
  91     stat_put(dict, "mem_swapped_in", dev->stats[VIRTIO_BALLOON_S_SWAP_IN]);
  92     stat_put(dict, "mem_swapped_out", dev->stats[VIRTIO_BALLOON_S_SWAP_OUT]);
  93     stat_put(dict, "major_page_faults", dev->stats[VIRTIO_BALLOON_S_MAJFLT]);
  94     stat_put(dict, "minor_page_faults", dev->stats[VIRTIO_BALLOON_S_MINFLT]);
  95     stat_put(dict, "free_mem", dev->stats[VIRTIO_BALLOON_S_MEMFREE]);
  96     stat_put(dict, "total_mem", dev->stats[VIRTIO_BALLOON_S_MEMTOT]);
  97 #endif
  98
  99     return QOBJECT(dict);
 100 }
 101
 102 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 103 {
 104     VirtIOBalloon *s = to_virtio_balloon(vdev);
 105     VirtQueueElement elem;
 106
 107     while (virtqueue_pop(vq, &elem)) {
 108         size_t offset = 0;
 109         uint32_t pfn;
 110
 111         while (iov_to_buf(elem.out_sg, elem.out_num, &pfn, offset, 4) == 4) {
 112             ram_addr_t pa;
 113             ram_addr_t addr;
 114
 115             pa = (ram_addr_t)ldl_p(&pfn) << VIRTIO_BALLOON_PFN_SHIFT;
 116             offset += 4;
 117
 118             addr = cpu_get_physical_page_desc(pa);
 119             if ((addr & ~TARGET_PAGE_MASK) != IO_MEM_RAM)
 120                 continue;
 121
 122             /* Using qemu_get_ram_ptr is bending the rules a bit, but
 123                should be OK because we only want a single page.  */
 124             balloon_page(qemu_get_ram_ptr(addr), !!(vq == s->dvq));
 125         }
 126
 127         virtqueue_push(vq, &elem, offset);
 128         virtio_notify(vdev, vq);
 129     }
 130 }
 131
 132 static void complete_stats_request(VirtIOBalloon *vb)
 133 {
 134     QObject *stats;
 135
 136     if (!vb->stats_opaque_callback_data)
 137         return;
 138
 139     stats = get_stats_qobject(vb);
 140     vb->stats_callback(vb->stats_opaque_callback_data, stats);
 141     qobject_decref(stats);
 142     vb->stats_opaque_callback_data = NULL;
 143     vb->stats_callback = NULL;
 144 }
 145
 146 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
 147 {
 148     VirtIOBalloon *s = DO_UPCAST(VirtIOBalloon, vdev, vdev);
 149     VirtQueueElement *elem = &s->stats_vq_elem;
 150     VirtIOBalloonStat stat;
 151     size_t offset = 0;
 152
 153     if (!virtqueue_pop(vq, elem)) {
 154         return;
 155     }
 156
 157     /* Initialize the stats to get rid of any stale values.  This is only
 158      * needed to handle the case where a guest supports fewer stats than it
 159      * used to (ie. it has booted into an old kernel).
 160      */
 161     reset_stats(s);
 162
 163     while (iov_to_buf(elem->out_sg, elem->out_num, &stat, offset, sizeof(stat))
 164            == sizeof(stat)) {
 165         uint16_t tag = tswap16(stat.tag);
 166         uint64_t val = tswap64(stat.val);
 167
 168         offset += sizeof(stat);
 169         if (tag < VIRTIO_BALLOON_S_NR)
 170             s->stats[tag] = val;
 171     }
 172     s->stats_vq_offset = offset;
 173
 174     complete_stats_request(s);
 175 }
 176
 177 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
 178 {
 179     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 180     struct virtio_balloon_config config;
 181
 182     config.num_pages = cpu_to_le32(dev->num_pages);
 183     config.actual = cpu_to_le32(dev->actual);
 184
 185     memcpy(config_data, &config, 8);
 186 }
 187
 188 static void virtio_balloon_set_config(VirtIODevice *vdev,
 189                                       const uint8_t *config_data)
 190 {
 191     VirtIOBalloon *dev = to_virtio_balloon(vdev);
 192     struct virtio_balloon_config config;
 193     memcpy(&config, config_data, 8);
 194     dev->actual = le32_to_cpu(config.actual);
 195 }
 196
 197 static uint32_t virtio_balloon_get_features(VirtIODevice *vdev, uint32_t f)
 198 {
 199     f |= (1 << VIRTIO_BALLOON_F_STATS_VQ);
 200     return f;
 201 }
 202
 203 static void virtio_balloon_to_target(void *opaque, ram_addr_t target,
 204                                      MonitorCompletion cb, void *cb_data)
 205 {
 206     VirtIOBalloon *dev = opaque;
 207
 208     if (target > ram_size)
 209         target = ram_size;
 210
 211     if (target) {
 212         dev->num_pages = (ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
 213         virtio_notify_config(&dev->vdev);
 214     } else {
 215         /* For now, only allow one request at a time.  This restriction can be
 216          * removed later by queueing callback and data pairs.
 217          */
 218         if (dev->stats_callback != NULL) {
 219             return;
 220         }
 221         dev->stats_callback = cb;
 222         dev->stats_opaque_callback_data = cb_data;
 223         if (ENABLE_GUEST_STATS && (dev->vdev.guest_features & (1 << VIRTIO_BALLOON_F_STATS_VQ))) {
 224             virtqueue_push(dev->svq, &dev->stats_vq_elem, dev->stats_vq_offset);
 225             virtio_notify(&dev->vdev, dev->svq);
 226         } else {
 227             /* Stats are not supported.  Clear out any stale values that might
 228              * have been set by a more featureful guest kernel.
 229              */
 230             reset_stats(dev);
 231             complete_stats_request(dev);
 232         }
 233     }
 234 }
 235
 236 static void virtio_balloon_save(QEMUFile *f, void *opaque)
 237 {
 238     VirtIOBalloon *s = opaque;
 239
 240     virtio_save(&s->vdev, f);
 241
 242     qemu_put_be32(f, s->num_pages);
 243     qemu_put_be32(f, s->actual);
 244 }
 245
 246 static int virtio_balloon_load(QEMUFile *f, void *opaque, int version_id)
 247 {
 248     VirtIOBalloon *s = opaque;
 249
 250     if (version_id != 1)
 251         return -EINVAL;
 252
 253     virtio_load(&s->vdev, f);
 254
 255     s->num_pages = qemu_get_be32(f);
 256     s->actual = qemu_get_be32(f);
 257     return 0;
 258 }
 259
 260 VirtIODevice *virtio_balloon_init(DeviceState *dev)
 261 {
 262     VirtIOBalloon *s;
 263
 264     s = (VirtIOBalloon *)virtio_common_init("virtio-balloon",
 265                                             VIRTIO_ID_BALLOON,
 266                                             8, sizeof(VirtIOBalloon));
 267
 268     s->vdev.get_config = virtio_balloon_get_config;
 269     s->vdev.set_config = virtio_balloon_set_config;
 270     s->vdev.get_features = virtio_balloon_get_features;
 271
 272     s->ivq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 273     s->dvq = virtio_add_queue(&s->vdev, 128, virtio_balloon_handle_output);
 274     s->svq = virtio_add_queue(&s->vdev, 128, virtio_balloon_receive_stats);
 275
 276     reset_stats(s);
 277     qemu_add_balloon_handler(virtio_balloon_to_target, s);
 278
 279     register_savevm(dev, "virtio-balloon", -1, 1,
 280                     virtio_balloon_save, virtio_balloon_load, s);
 281
 282     return &s->vdev;
 283 }