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 "hw/mem/nvdimm.h"
  24 #include "qapi/error.h"
  25 #include "qemu/config-file.h"
  26 #include "qapi/visitor.h"
  27 #include "qemu/range.h"
  28 #include "sysemu/numa.h"
  29 #include "sysemu/kvm.h"
  30 #include "trace.h"
  31 #include "hw/virtio/vhost.h"
  32
  33 typedef struct pc_dimms_capacity {
  34      uint64_t size;
  35      Error    **errp;
  36 } pc_dimms_capacity;
  37
  38 void pc_dimm_memory_plug(DeviceState *dev, MemoryHotplugState *hpms,
  39                          MemoryRegion *mr, uint64_t align, Error **errp)
  40 {
  41     int slot;
  42     MachineState *machine = MACHINE(qdev_get_machine());
  43     PCDIMMDevice *dimm = PC_DIMM(dev);
  44     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
  45     MemoryRegion *vmstate_mr = ddc->get_vmstate_memory_region(dimm);
  46     Error *local_err = NULL;
  47     uint64_t existing_dimms_capacity = 0;
  48     uint64_t addr;
  49
  50     addr = object_property_get_uint(OBJECT(dimm),
  51                                     PC_DIMM_ADDR_PROP, &local_err);
  52     if (local_err) {
  53         goto out;
  54     }
  55
  56     addr = pc_dimm_get_free_addr(hpms->base,
  57                                  memory_region_size(&hpms->mr),
  58                                  !addr ? NULL : &addr, align,
  59                                  memory_region_size(mr), &local_err);
  60     if (local_err) {
  61         goto out;
  62     }
  63
  64     existing_dimms_capacity = pc_existing_dimms_capacity(&local_err);
  65     if (local_err) {
  66         goto out;
  67     }
  68
  69     if (existing_dimms_capacity + memory_region_size(mr) >
  70         machine->maxram_size - machine->ram_size) {
  71         error_setg(&local_err, "not enough space, currently 0x%" PRIx64
  72                    " in use of total hot pluggable 0x" RAM_ADDR_FMT,
  73                    existing_dimms_capacity,
  74                    machine->maxram_size - machine->ram_size);
  75         goto out;
  76     }
  77
  78     object_property_set_uint(OBJECT(dev), addr, PC_DIMM_ADDR_PROP, &local_err);
  79     if (local_err) {
  80         goto out;
  81     }
  82     trace_mhp_pc_dimm_assigned_address(addr);
  83
  84     slot = object_property_get_int(OBJECT(dev), PC_DIMM_SLOT_PROP, &local_err);
  85     if (local_err) {
  86         goto out;
  87     }
  88
  89     slot = pc_dimm_get_free_slot(slot == PC_DIMM_UNASSIGNED_SLOT ? NULL : &slot,
  90                                  machine->ram_slots, &local_err);
  91     if (local_err) {
  92         goto out;
  93     }
  94     object_property_set_int(OBJECT(dev), slot, PC_DIMM_SLOT_PROP, &local_err);
  95     if (local_err) {
  96         goto out;
  97     }
  98     trace_mhp_pc_dimm_assigned_slot(slot);
  99
 100     if (kvm_enabled() && !kvm_has_free_slot(machine)) {
 101         error_setg(&local_err, "hypervisor has no free memory slots left");
 102         goto out;
 103     }
 104
 105     if (!vhost_has_free_slot()) {
 106         error_setg(&local_err, "a used vhost backend has no free"
 107                                " memory slots left");
 108         goto out;
 109     }
 110
 111     memory_region_add_subregion(&hpms->mr, addr - hpms->base, mr);
 112     vmstate_register_ram(vmstate_mr, dev);
 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     memory_region_del_subregion(&hpms->mr, mr);
 126     vmstate_unregister_ram(vmstate_mr, dev);
 127 }
 128
 129 static int pc_existing_dimms_capacity_internal(Object *obj, void *opaque)
 130 {
 131     pc_dimms_capacity *cap = opaque;
 132     uint64_t *size = &cap->size;
 133
 134     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 135         DeviceState *dev = DEVICE(obj);
 136
 137         if (dev->realized) {
 138             (*size) += object_property_get_uint(obj, PC_DIMM_SIZE_PROP,
 139                 cap->errp);
 140         }
 141
 142         if (cap->errp && *cap->errp) {
 143             return 1;
 144         }
 145     }
 146     object_child_foreach(obj, pc_existing_dimms_capacity_internal, opaque);
 147     return 0;
 148 }
 149
 150 uint64_t pc_existing_dimms_capacity(Error **errp)
 151 {
 152     pc_dimms_capacity cap;
 153
 154     cap.size = 0;
 155     cap.errp = errp;
 156
 157     pc_existing_dimms_capacity_internal(qdev_get_machine(), &cap);
 158     return cap.size;
 159 }
 160
 161 uint64_t get_plugged_memory_size(void)
 162 {
 163     return pc_existing_dimms_capacity(&error_abort);
 164 }
 165
 166 static int pc_dimm_slot2bitmap(Object *obj, void *opaque)
 167 {
 168     unsigned long *bitmap = opaque;
 169
 170     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 171         DeviceState *dev = DEVICE(obj);
 172         if (dev->realized) { /* count only realized DIMMs */
 173             PCDIMMDevice *d = PC_DIMM(obj);
 174             set_bit(d->slot, bitmap);
 175         }
 176     }
 177
 178     object_child_foreach(obj, pc_dimm_slot2bitmap, opaque);
 179     return 0;
 180 }
 181
 182 int pc_dimm_get_free_slot(const int *hint, int max_slots, Error **errp)
 183 {
 184     unsigned long *bitmap = bitmap_new(max_slots);
 185     int slot = 0;
 186
 187     object_child_foreach(qdev_get_machine(), pc_dimm_slot2bitmap, bitmap);
 188
 189     /* check if requested slot is not occupied */
 190     if (hint) {
 191         if (*hint >= max_slots) {
 192             error_setg(errp, "invalid slot# %d, should be less than %d",
 193                        *hint, max_slots);
 194         } else if (!test_bit(*hint, bitmap)) {
 195             slot = *hint;
 196         } else {
 197             error_setg(errp, "slot %d is busy", *hint);
 198         }
 199         goto out;
 200     }
 201
 202     /* search for free slot */
 203     slot = find_first_zero_bit(bitmap, max_slots);
 204     if (slot == max_slots) {
 205         error_setg(errp, "no free slots available");
 206     }
 207 out:
 208     g_free(bitmap);
 209     return slot;
 210 }
 211
 212 static gint pc_dimm_addr_sort(gconstpointer a, gconstpointer b)
 213 {
 214     PCDIMMDevice *x = PC_DIMM(a);
 215     PCDIMMDevice *y = PC_DIMM(b);
 216     Int128 diff = int128_sub(int128_make64(x->addr), int128_make64(y->addr));
 217
 218     if (int128_lt(diff, int128_zero())) {
 219         return -1;
 220     } else if (int128_gt(diff, int128_zero())) {
 221         return 1;
 222     }
 223     return 0;
 224 }
 225
 226 static int pc_dimm_built_list(Object *obj, void *opaque)
 227 {
 228     GSList **list = opaque;
 229
 230     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 231         DeviceState *dev = DEVICE(obj);
 232         if (dev->realized) { /* only realized DIMMs matter */
 233             *list = g_slist_insert_sorted(*list, dev, pc_dimm_addr_sort);
 234         }
 235     }
 236
 237     object_child_foreach(obj, pc_dimm_built_list, opaque);
 238     return 0;
 239 }
 240
 241 MemoryDeviceInfoList *qmp_pc_dimm_device_list(void)
 242 {
 243     GSList *dimms = NULL, *item;
 244     MemoryDeviceInfoList *list = NULL, *prev = NULL;
 245
 246     object_child_foreach(qdev_get_machine(), pc_dimm_built_list, &dimms);
 247
 248     for (item = dimms; item; item = g_slist_next(item)) {
 249         PCDIMMDevice *dimm = PC_DIMM(item->data);
 250         Object *obj = OBJECT(dimm);
 251         MemoryDeviceInfoList *elem = g_new0(MemoryDeviceInfoList, 1);
 252         MemoryDeviceInfo *info = g_new0(MemoryDeviceInfo, 1);
 253         PCDIMMDeviceInfo *di = g_new0(PCDIMMDeviceInfo, 1);
 254         bool is_nvdimm = object_dynamic_cast(obj, TYPE_NVDIMM);
 255         DeviceClass *dc = DEVICE_GET_CLASS(obj);
 256         DeviceState *dev = DEVICE(obj);
 257
 258         if (dev->id) {
 259             di->has_id = true;
 260             di->id = g_strdup(dev->id);
 261         }
 262         di->hotplugged = dev->hotplugged;
 263         di->hotpluggable = dc->hotpluggable;
 264         di->addr = dimm->addr;
 265         di->slot = dimm->slot;
 266         di->node = dimm->node;
 267         di->size = object_property_get_uint(obj, PC_DIMM_SIZE_PROP, NULL);
 268         di->memdev = object_get_canonical_path(OBJECT(dimm->hostmem));
 269
 270         if (!is_nvdimm) {
 271             info->u.dimm.data = di;
 272             info->type = MEMORY_DEVICE_INFO_KIND_DIMM;
 273         } else {
 274             info->u.nvdimm.data = di;
 275             info->type = MEMORY_DEVICE_INFO_KIND_NVDIMM;
 276         }
 277         elem->value = info;
 278         elem->next = NULL;
 279         if (prev) {
 280             prev->next = elem;
 281         } else {
 282             list = elem;
 283         }
 284         prev = elem;
 285     }
 286
 287     g_slist_free(dimms);
 288
 289     return list;
 290 }
 291
 292 uint64_t pc_dimm_get_free_addr(uint64_t address_space_start,
 293                                uint64_t address_space_size,
 294                                uint64_t *hint, uint64_t align, uint64_t size,
 295                                Error **errp)
 296 {
 297     GSList *list = NULL, *item;
 298     uint64_t new_addr, ret = 0;
 299     uint64_t address_space_end = address_space_start + address_space_size;
 300
 301     g_assert(QEMU_ALIGN_UP(address_space_start, align) == address_space_start);
 302
 303     if (!address_space_size) {
 304         error_setg(errp, "memory hotplug is not enabled, "
 305                          "please add maxmem option");
 306         goto out;
 307     }
 308
 309     if (hint && QEMU_ALIGN_UP(*hint, align) != *hint) {
 310         error_setg(errp, "address must be aligned to 0x%" PRIx64 " bytes",
 311                    align);
 312         goto out;
 313     }
 314
 315     if (QEMU_ALIGN_UP(size, align) != size) {
 316         error_setg(errp, "backend memory size must be multiple of 0x%"
 317                    PRIx64, align);
 318         goto out;
 319     }
 320
 321     assert(address_space_end > address_space_start);
 322     object_child_foreach(qdev_get_machine(), pc_dimm_built_list, &list);
 323
 324     if (hint) {
 325         new_addr = *hint;
 326     } else {
 327         new_addr = address_space_start;
 328     }
 329
 330     /* find address range that will fit new DIMM */
 331     for (item = list; item; item = g_slist_next(item)) {
 332         PCDIMMDevice *dimm = item->data;
 333         uint64_t dimm_size = object_property_get_uint(OBJECT(dimm),
 334                                                       PC_DIMM_SIZE_PROP,
 335                                                       errp);
 336         if (errp && *errp) {
 337             goto out;
 338         }
 339
 340         if (ranges_overlap(dimm->addr, dimm_size, new_addr, size)) {
 341             if (hint) {
 342                 DeviceState *d = DEVICE(dimm);
 343                 error_setg(errp, "address range conflicts with '%s'", d->id);
 344                 goto out;
 345             }
 346             new_addr = QEMU_ALIGN_UP(dimm->addr + dimm_size, align);
 347         }
 348     }
 349     ret = new_addr;
 350
 351     if (new_addr < address_space_start) {
 352         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 353                    "] at 0x%" PRIx64, new_addr, size, address_space_start);
 354     } else if ((new_addr + size) > address_space_end) {
 355         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 356                    "] beyond 0x%" PRIx64, new_addr, size, address_space_end);
 357     }
 358
 359 out:
 360     g_slist_free(list);
 361     return ret;
 362 }
 363
 364 static Property pc_dimm_properties[] = {
 365     DEFINE_PROP_UINT64(PC_DIMM_ADDR_PROP, PCDIMMDevice, addr, 0),
 366     DEFINE_PROP_UINT32(PC_DIMM_NODE_PROP, PCDIMMDevice, node, 0),
 367     DEFINE_PROP_INT32(PC_DIMM_SLOT_PROP, PCDIMMDevice, slot,
 368                       PC_DIMM_UNASSIGNED_SLOT),
 369     DEFINE_PROP_LINK(PC_DIMM_MEMDEV_PROP, PCDIMMDevice, hostmem,
 370                      TYPE_MEMORY_BACKEND, HostMemoryBackend *),
 371     DEFINE_PROP_END_OF_LIST(),
 372 };
 373
 374 static void pc_dimm_get_size(Object *obj, Visitor *v, const char *name,
 375                              void *opaque, Error **errp)
 376 {
 377     uint64_t value;
 378     MemoryRegion *mr;
 379     PCDIMMDevice *dimm = PC_DIMM(obj);
 380     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(obj);
 381
 382     mr = ddc->get_memory_region(dimm, errp);
 383     if (!mr) {
 384         return;
 385     }
 386     value = memory_region_size(mr);
 387
 388     visit_type_uint64(v, name, &value, errp);
 389 }
 390
 391 static void pc_dimm_init(Object *obj)
 392 {
 393     object_property_add(obj, PC_DIMM_SIZE_PROP, "uint64", pc_dimm_get_size,
 394                         NULL, NULL, NULL, &error_abort);
 395 }
 396
 397 static void pc_dimm_realize(DeviceState *dev, Error **errp)
 398 {
 399     PCDIMMDevice *dimm = PC_DIMM(dev);
 400     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
 401
 402     if (!dimm->hostmem) {
 403         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property is not set");
 404         return;
 405     } else if (host_memory_backend_is_mapped(dimm->hostmem)) {
 406         char *path = object_get_canonical_path_component(OBJECT(dimm->hostmem));
 407         error_setg(errp, "can't use already busy memdev: %s", path);
 408         g_free(path);
 409         return;
 410     }
 411     if (((nb_numa_nodes > 0) && (dimm->node >= nb_numa_nodes)) ||
 412         (!nb_numa_nodes && dimm->node)) {
 413         error_setg(errp, "'DIMM property " PC_DIMM_NODE_PROP " has value %"
 414                    PRIu32 "' which exceeds the number of numa nodes: %d",
 415                    dimm->node, nb_numa_nodes ? nb_numa_nodes : 1);
 416         return;
 417     }
 418
 419     if (ddc->realize) {
 420         ddc->realize(dimm, errp);
 421     }
 422
 423     host_memory_backend_set_mapped(dimm->hostmem, true);
 424 }
 425
 426 static void pc_dimm_unrealize(DeviceState *dev, Error **errp)
 427 {
 428     PCDIMMDevice *dimm = PC_DIMM(dev);
 429
 430     host_memory_backend_set_mapped(dimm->hostmem, false);
 431 }
 432
 433 static MemoryRegion *pc_dimm_get_memory_region(PCDIMMDevice *dimm, Error **errp)
 434 {
 435     if (!dimm->hostmem) {
 436         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property must be set");
 437         return NULL;
 438     }
 439
 440     return host_memory_backend_get_memory(dimm->hostmem, errp);
 441 }
 442
 443 static MemoryRegion *pc_dimm_get_vmstate_memory_region(PCDIMMDevice *dimm)
 444 {
 445     return host_memory_backend_get_memory(dimm->hostmem, &error_abort);
 446 }
 447
 448 static void pc_dimm_class_init(ObjectClass *oc, void *data)
 449 {
 450     DeviceClass *dc = DEVICE_CLASS(oc);
 451     PCDIMMDeviceClass *ddc = PC_DIMM_CLASS(oc);
 452
 453     dc->realize = pc_dimm_realize;
 454     dc->unrealize = pc_dimm_unrealize;
 455     dc->props = pc_dimm_properties;
 456     dc->desc = "DIMM memory module";
 457
 458     ddc->get_memory_region = pc_dimm_get_memory_region;
 459     ddc->get_vmstate_memory_region = pc_dimm_get_vmstate_memory_region;
 460 }
 461
 462 static TypeInfo pc_dimm_info = {
 463     .name          = TYPE_PC_DIMM,
 464     .parent        = TYPE_DEVICE,
 465     .instance_size = sizeof(PCDIMMDevice),
 466     .instance_init = pc_dimm_init,
 467     .class_init    = pc_dimm_class_init,
 468     .class_size    = sizeof(PCDIMMDeviceClass),
 469 };
 470
 471 static void pc_dimm_register_types(void)
 472 {
 473     type_register_static(&pc_dimm_info);
 474 }
 475
 476 type_init(pc_dimm_register_types)