hw/mem/pc-dimm.c

   1 /*
   2  * Dimm device for Memory Hotplug
   3  *
   4  * Copyright ProfitBricks GmbH 2012
   5  * Copyright (C) 2014 Red Hat Inc
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, see <http://www.gnu.org/licenses/>
  19  */
  20
  21 #include "qemu/osdep.h"
  22 #include "hw/mem/pc-dimm.h"
  23 #include "qapi/error.h"
  24 #include "qemu/config-file.h"
  25 #include "qapi/visitor.h"
  26 #include "qemu/range.h"
  27 #include "sysemu/numa.h"
  28 #include "sysemu/kvm.h"
  29 #include "trace.h"
  30 #include "hw/virtio/vhost.h"
  31
  32 typedef struct pc_dimms_capacity {
  33      uint64_t size;
  34      Error    **errp;
  35 } pc_dimms_capacity;
  36
  37 void pc_dimm_memory_plug(DeviceState *dev, MemoryHotplugState *hpms,
  38                          MemoryRegion *mr, uint64_t align, Error **errp)
  39 {
  40     int slot;
  41     MachineState *machine = MACHINE(qdev_get_machine());
  42     PCDIMMDevice *dimm = PC_DIMM(dev);
  43     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
  44     MemoryRegion *vmstate_mr = ddc->get_vmstate_memory_region(dimm);
  45     Error *local_err = NULL;
  46     uint64_t existing_dimms_capacity = 0;
  47     uint64_t addr;
  48
  49     addr = object_property_get_uint(OBJECT(dimm),
  50                                     PC_DIMM_ADDR_PROP, &local_err);
  51     if (local_err) {
  52         goto out;
  53     }
  54
  55     addr = pc_dimm_get_free_addr(hpms->base,
  56                                  memory_region_size(&hpms->mr),
  57                                  !addr ? NULL : &addr, align,
  58                                  memory_region_size(mr), &local_err);
  59     if (local_err) {
  60         goto out;
  61     }
  62
  63     existing_dimms_capacity = pc_existing_dimms_capacity(&local_err);
  64     if (local_err) {
  65         goto out;
  66     }
  67
  68     if (existing_dimms_capacity + memory_region_size(mr) >
  69         machine->maxram_size - machine->ram_size) {
  70         error_setg(&local_err, "not enough space, currently 0x%" PRIx64
  71                    " in use of total hot pluggable 0x" RAM_ADDR_FMT,
  72                    existing_dimms_capacity,
  73                    machine->maxram_size - machine->ram_size);
  74         goto out;
  75     }
  76
  77     object_property_set_uint(OBJECT(dev), addr, PC_DIMM_ADDR_PROP, &local_err);
  78     if (local_err) {
  79         goto out;
  80     }
  81     trace_mhp_pc_dimm_assigned_address(addr);
  82
  83     slot = object_property_get_int(OBJECT(dev), PC_DIMM_SLOT_PROP, &local_err);
  84     if (local_err) {
  85         goto out;
  86     }
  87
  88     slot = pc_dimm_get_free_slot(slot == PC_DIMM_UNASSIGNED_SLOT ? NULL : &slot,
  89                                  machine->ram_slots, &local_err);
  90     if (local_err) {
  91         goto out;
  92     }
  93     object_property_set_int(OBJECT(dev), slot, PC_DIMM_SLOT_PROP, &local_err);
  94     if (local_err) {
  95         goto out;
  96     }
  97     trace_mhp_pc_dimm_assigned_slot(slot);
  98
  99     if (kvm_enabled() && !kvm_has_free_slot(machine)) {
 100         error_setg(&local_err, "hypervisor has no free memory slots left");
 101         goto out;
 102     }
 103
 104     if (!vhost_has_free_slot()) {
 105         error_setg(&local_err, "a used vhost backend has no free"
 106                                " memory slots left");
 107         goto out;
 108     }
 109
 110     memory_region_add_subregion(&hpms->mr, addr - hpms->base, mr);
 111     vmstate_register_ram(vmstate_mr, dev);
 112     numa_set_mem_node_id(addr, memory_region_size(mr), dimm->node);
 113
 114 out:
 115     error_propagate(errp, local_err);
 116 }
 117
 118 void pc_dimm_memory_unplug(DeviceState *dev, MemoryHotplugState *hpms,
 119                            MemoryRegion *mr)
 120 {
 121     PCDIMMDevice *dimm = PC_DIMM(dev);
 122     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
 123     MemoryRegion *vmstate_mr = ddc->get_vmstate_memory_region(dimm);
 124
 125     numa_unset_mem_node_id(dimm->addr, memory_region_size(mr), dimm->node);
 126     memory_region_del_subregion(&hpms->mr, mr);
 127     vmstate_unregister_ram(vmstate_mr, dev);
 128 }
 129
 130 static int pc_existing_dimms_capacity_internal(Object *obj, void *opaque)
 131 {
 132     pc_dimms_capacity *cap = opaque;
 133     uint64_t *size = &cap->size;
 134
 135     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 136         DeviceState *dev = DEVICE(obj);
 137
 138         if (dev->realized) {
 139             (*size) += object_property_get_uint(obj, PC_DIMM_SIZE_PROP,
 140                 cap->errp);
 141         }
 142
 143         if (cap->errp && *cap->errp) {
 144             return 1;
 145         }
 146     }
 147     object_child_foreach(obj, pc_existing_dimms_capacity_internal, opaque);
 148     return 0;
 149 }
 150
 151 uint64_t pc_existing_dimms_capacity(Error **errp)
 152 {
 153     pc_dimms_capacity cap;
 154
 155     cap.size = 0;
 156     cap.errp = errp;
 157
 158     pc_existing_dimms_capacity_internal(qdev_get_machine(), &cap);
 159     return cap.size;
 160 }
 161
 162 int qmp_pc_dimm_device_list(Object *obj, void *opaque)
 163 {
 164     MemoryDeviceInfoList ***prev = opaque;
 165
 166     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 167         DeviceState *dev = DEVICE(obj);
 168
 169         if (dev->realized) {
 170             MemoryDeviceInfoList *elem = g_new0(MemoryDeviceInfoList, 1);
 171             MemoryDeviceInfo *info = g_new0(MemoryDeviceInfo, 1);
 172             PCDIMMDeviceInfo *di = g_new0(PCDIMMDeviceInfo, 1);
 173             DeviceClass *dc = DEVICE_GET_CLASS(obj);
 174             PCDIMMDevice *dimm = PC_DIMM(obj);
 175
 176             if (dev->id) {
 177                 di->has_id = true;
 178                 di->id = g_strdup(dev->id);
 179             }
 180             di->hotplugged = dev->hotplugged;
 181             di->hotpluggable = dc->hotpluggable;
 182             di->addr = dimm->addr;
 183             di->slot = dimm->slot;
 184             di->node = dimm->node;
 185             di->size = object_property_get_uint(OBJECT(dimm), PC_DIMM_SIZE_PROP,
 186                                                 NULL);
 187             di->memdev = object_get_canonical_path(OBJECT(dimm->hostmem));
 188
 189             info->u.dimm.data = di;
 190             elem->value = info;
 191             elem->next = NULL;
 192             **prev = elem;
 193             *prev = &elem->next;
 194         }
 195     }
 196
 197     object_child_foreach(obj, qmp_pc_dimm_device_list, opaque);
 198     return 0;
 199 }
 200
 201 static int pc_dimm_slot2bitmap(Object *obj, void *opaque)
 202 {
 203     unsigned long *bitmap = opaque;
 204
 205     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 206         DeviceState *dev = DEVICE(obj);
 207         if (dev->realized) { /* count only realized DIMMs */
 208             PCDIMMDevice *d = PC_DIMM(obj);
 209             set_bit(d->slot, bitmap);
 210         }
 211     }
 212
 213     object_child_foreach(obj, pc_dimm_slot2bitmap, opaque);
 214     return 0;
 215 }
 216
 217 int pc_dimm_get_free_slot(const int *hint, int max_slots, Error **errp)
 218 {
 219     unsigned long *bitmap = bitmap_new(max_slots);
 220     int slot = 0;
 221
 222     object_child_foreach(qdev_get_machine(), pc_dimm_slot2bitmap, bitmap);
 223
 224     /* check if requested slot is not occupied */
 225     if (hint) {
 226         if (*hint >= max_slots) {
 227             error_setg(errp, "invalid slot# %d, should be less than %d",
 228                        *hint, max_slots);
 229         } else if (!test_bit(*hint, bitmap)) {
 230             slot = *hint;
 231         } else {
 232             error_setg(errp, "slot %d is busy", *hint);
 233         }
 234         goto out;
 235     }
 236
 237     /* search for free slot */
 238     slot = find_first_zero_bit(bitmap, max_slots);
 239     if (slot == max_slots) {
 240         error_setg(errp, "no free slots available");
 241     }
 242 out:
 243     g_free(bitmap);
 244     return slot;
 245 }
 246
 247 static gint pc_dimm_addr_sort(gconstpointer a, gconstpointer b)
 248 {
 249     PCDIMMDevice *x = PC_DIMM(a);
 250     PCDIMMDevice *y = PC_DIMM(b);
 251     Int128 diff = int128_sub(int128_make64(x->addr), int128_make64(y->addr));
 252
 253     if (int128_lt(diff, int128_zero())) {
 254         return -1;
 255     } else if (int128_gt(diff, int128_zero())) {
 256         return 1;
 257     }
 258     return 0;
 259 }
 260
 261 static int pc_dimm_built_list(Object *obj, void *opaque)
 262 {
 263     GSList **list = opaque;
 264
 265     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 266         DeviceState *dev = DEVICE(obj);
 267         if (dev->realized) { /* only realized DIMMs matter */
 268             *list = g_slist_insert_sorted(*list, dev, pc_dimm_addr_sort);
 269         }
 270     }
 271
 272     object_child_foreach(obj, pc_dimm_built_list, opaque);
 273     return 0;
 274 }
 275
 276 uint64_t pc_dimm_get_free_addr(uint64_t address_space_start,
 277                                uint64_t address_space_size,
 278                                uint64_t *hint, uint64_t align, uint64_t size,
 279                                Error **errp)
 280 {
 281     GSList *list = NULL, *item;
 282     uint64_t new_addr, ret = 0;
 283     uint64_t address_space_end = address_space_start + address_space_size;
 284
 285     g_assert(QEMU_ALIGN_UP(address_space_start, align) == address_space_start);
 286
 287     if (!address_space_size) {
 288         error_setg(errp, "memory hotplug is not enabled, "
 289                          "please add maxmem option");
 290         goto out;
 291     }
 292
 293     if (hint && QEMU_ALIGN_UP(*hint, align) != *hint) {
 294         error_setg(errp, "address must be aligned to 0x%" PRIx64 " bytes",
 295                    align);
 296         goto out;
 297     }
 298
 299     if (QEMU_ALIGN_UP(size, align) != size) {
 300         error_setg(errp, "backend memory size must be multiple of 0x%"
 301                    PRIx64, align);
 302         goto out;
 303     }
 304
 305     assert(address_space_end > address_space_start);
 306     object_child_foreach(qdev_get_machine(), pc_dimm_built_list, &list);
 307
 308     if (hint) {
 309         new_addr = *hint;
 310     } else {
 311         new_addr = address_space_start;
 312     }
 313
 314     /* find address range that will fit new DIMM */
 315     for (item = list; item; item = g_slist_next(item)) {
 316         PCDIMMDevice *dimm = item->data;
 317         uint64_t dimm_size = object_property_get_uint(OBJECT(dimm),
 318                                                       PC_DIMM_SIZE_PROP,
 319                                                       errp);
 320         if (errp && *errp) {
 321             goto out;
 322         }
 323
 324         if (ranges_overlap(dimm->addr, dimm_size, new_addr, size)) {
 325             if (hint) {
 326                 DeviceState *d = DEVICE(dimm);
 327                 error_setg(errp, "address range conflicts with '%s'", d->id);
 328                 goto out;
 329             }
 330             new_addr = QEMU_ALIGN_UP(dimm->addr + dimm_size, align);
 331         }
 332     }
 333     ret = new_addr;
 334
 335     if (new_addr < address_space_start) {
 336         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 337                    "] at 0x%" PRIx64, new_addr, size, address_space_start);
 338     } else if ((new_addr + size) > address_space_end) {
 339         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 340                    "] beyond 0x%" PRIx64, new_addr, size, address_space_end);
 341     }
 342
 343 out:
 344     g_slist_free(list);
 345     return ret;
 346 }
 347
 348 static Property pc_dimm_properties[] = {
 349     DEFINE_PROP_UINT64(PC_DIMM_ADDR_PROP, PCDIMMDevice, addr, 0),
 350     DEFINE_PROP_UINT32(PC_DIMM_NODE_PROP, PCDIMMDevice, node, 0),
 351     DEFINE_PROP_INT32(PC_DIMM_SLOT_PROP, PCDIMMDevice, slot,
 352                       PC_DIMM_UNASSIGNED_SLOT),
 353     DEFINE_PROP_LINK(PC_DIMM_MEMDEV_PROP, PCDIMMDevice, hostmem,
 354                      TYPE_MEMORY_BACKEND, HostMemoryBackend *),
 355     DEFINE_PROP_END_OF_LIST(),
 356 };
 357
 358 static void pc_dimm_get_size(Object *obj, Visitor *v, const char *name,
 359                              void *opaque, Error **errp)
 360 {
 361     uint64_t value;
 362     MemoryRegion *mr;
 363     PCDIMMDevice *dimm = PC_DIMM(obj);
 364     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(obj);
 365
 366     mr = ddc->get_memory_region(dimm, errp);
 367     if (!mr) {
 368         return;
 369     }
 370     value = memory_region_size(mr);
 371
 372     visit_type_uint64(v, name, &value, errp);
 373 }
 374
 375 static void pc_dimm_init(Object *obj)
 376 {
 377     object_property_add(obj, PC_DIMM_SIZE_PROP, "uint64", pc_dimm_get_size,
 378                         NULL, NULL, NULL, &error_abort);
 379 }
 380
 381 static void pc_dimm_realize(DeviceState *dev, Error **errp)
 382 {
 383     PCDIMMDevice *dimm = PC_DIMM(dev);
 384     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
 385
 386     if (!dimm->hostmem) {
 387         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property is not set");
 388         return;
 389     } else if (host_memory_backend_is_mapped(dimm->hostmem)) {
 390         char *path = object_get_canonical_path_component(OBJECT(dimm->hostmem));
 391         error_setg(errp, "can't use already busy memdev: %s", path);
 392         g_free(path);
 393         return;
 394     }
 395     if (((nb_numa_nodes > 0) && (dimm->node >= nb_numa_nodes)) ||
 396         (!nb_numa_nodes && dimm->node)) {
 397         error_setg(errp, "'DIMM property " PC_DIMM_NODE_PROP " has value %"
 398                    PRIu32 "' which exceeds the number of numa nodes: %d",
 399                    dimm->node, nb_numa_nodes ? nb_numa_nodes : 1);
 400         return;
 401     }
 402
 403     if (ddc->realize) {
 404         ddc->realize(dimm, errp);
 405     }
 406
 407     host_memory_backend_set_mapped(dimm->hostmem, true);
 408 }
 409
 410 static void pc_dimm_unrealize(DeviceState *dev, Error **errp)
 411 {
 412     PCDIMMDevice *dimm = PC_DIMM(dev);
 413
 414     host_memory_backend_set_mapped(dimm->hostmem, false);
 415 }
 416
 417 static MemoryRegion *pc_dimm_get_memory_region(PCDIMMDevice *dimm, Error **errp)
 418 {
 419     if (!dimm->hostmem) {
 420         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property must be set");
 421         return NULL;
 422     }
 423
 424     return host_memory_backend_get_memory(dimm->hostmem, errp);
 425 }
 426
 427 static MemoryRegion *pc_dimm_get_vmstate_memory_region(PCDIMMDevice *dimm)
 428 {
 429     return host_memory_backend_get_memory(dimm->hostmem, &error_abort);
 430 }
 431
 432 static void pc_dimm_class_init(ObjectClass *oc, void *data)
 433 {
 434     DeviceClass *dc = DEVICE_CLASS(oc);
 435     PCDIMMDeviceClass *ddc = PC_DIMM_CLASS(oc);
 436
 437     dc->realize = pc_dimm_realize;
 438     dc->unrealize = pc_dimm_unrealize;
 439     dc->props = pc_dimm_properties;
 440     dc->desc = "DIMM memory module";
 441
 442     ddc->get_memory_region = pc_dimm_get_memory_region;
 443     ddc->get_vmstate_memory_region = pc_dimm_get_vmstate_memory_region;
 444 }
 445
 446 static TypeInfo pc_dimm_info = {
 447     .name          = TYPE_PC_DIMM,
 448     .parent        = TYPE_DEVICE,
 449     .instance_size = sizeof(PCDIMMDevice),
 450     .instance_init = pc_dimm_init,
 451     .class_init    = pc_dimm_class_init,
 452     .class_size    = sizeof(PCDIMMDeviceClass),
 453 };
 454
 455 static void pc_dimm_register_types(void)
 456 {
 457     type_register_static(&pc_dimm_info);
 458 }
 459
 460 type_init(pc_dimm_register_types)