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 "hw/mem/pc-dimm.h"
  22 #include "qemu/config-file.h"
  23 #include "qapi/visitor.h"
  24 #include "qemu/range.h"
  25 #include "sysemu/numa.h"
  26 #include "sysemu/kvm.h"
  27 #include "trace.h"
  28 #include "hw/virtio/vhost.h"
  29
  30 typedef struct pc_dimms_capacity {
  31      uint64_t size;
  32      Error    **errp;
  33 } pc_dimms_capacity;
  34
  35 void pc_dimm_memory_plug(DeviceState *dev, MemoryHotplugState *hpms,
  36                          MemoryRegion *mr, uint64_t align, Error **errp)
  37 {
  38     int slot;
  39     MachineState *machine = MACHINE(qdev_get_machine());
  40     PCDIMMDevice *dimm = PC_DIMM(dev);
  41     Error *local_err = NULL;
  42     uint64_t existing_dimms_capacity = 0;
  43     uint64_t addr;
  44
  45     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
  46     if (local_err) {
  47         goto out;
  48     }
  49
  50     addr = pc_dimm_get_free_addr(hpms->base,
  51                                  memory_region_size(&hpms->mr),
  52                                  !addr ? NULL : &addr, align,
  53                                  memory_region_size(mr), &local_err);
  54     if (local_err) {
  55         goto out;
  56     }
  57
  58     existing_dimms_capacity = pc_existing_dimms_capacity(&local_err);
  59     if (local_err) {
  60         goto out;
  61     }
  62
  63     if (existing_dimms_capacity + memory_region_size(mr) >
  64         machine->maxram_size - machine->ram_size) {
  65         error_setg(&local_err, "not enough space, currently 0x%" PRIx64
  66                    " in use of total hot pluggable 0x" RAM_ADDR_FMT,
  67                    existing_dimms_capacity,
  68                    machine->maxram_size - machine->ram_size);
  69         goto out;
  70     }
  71
  72     object_property_set_int(OBJECT(dev), addr, PC_DIMM_ADDR_PROP, &local_err);
  73     if (local_err) {
  74         goto out;
  75     }
  76     trace_mhp_pc_dimm_assigned_address(addr);
  77
  78     slot = object_property_get_int(OBJECT(dev), PC_DIMM_SLOT_PROP, &local_err);
  79     if (local_err) {
  80         goto out;
  81     }
  82
  83     slot = pc_dimm_get_free_slot(slot == PC_DIMM_UNASSIGNED_SLOT ? NULL : &slot,
  84                                  machine->ram_slots, &local_err);
  85     if (local_err) {
  86         goto out;
  87     }
  88     object_property_set_int(OBJECT(dev), slot, PC_DIMM_SLOT_PROP, &local_err);
  89     if (local_err) {
  90         goto out;
  91     }
  92     trace_mhp_pc_dimm_assigned_slot(slot);
  93
  94     if (kvm_enabled() && !kvm_has_free_slot(machine)) {
  95         error_setg(&local_err, "hypervisor has no free memory slots left");
  96         goto out;
  97     }
  98
  99     if (!vhost_has_free_slot()) {
 100         error_setg(&local_err, "a used vhost backend has no free"
 101                                " memory slots left");
 102         goto out;
 103     }
 104
 105     memory_region_add_subregion(&hpms->mr, addr - hpms->base, mr);
 106     vmstate_register_ram(mr, dev);
 107     numa_set_mem_node_id(addr, memory_region_size(mr), dimm->node);
 108
 109 out:
 110     error_propagate(errp, local_err);
 111 }
 112
 113 void pc_dimm_memory_unplug(DeviceState *dev, MemoryHotplugState *hpms,
 114                            MemoryRegion *mr)
 115 {
 116     PCDIMMDevice *dimm = PC_DIMM(dev);
 117
 118     numa_unset_mem_node_id(dimm->addr, memory_region_size(mr), dimm->node);
 119     memory_region_del_subregion(&hpms->mr, mr);
 120     vmstate_unregister_ram(mr, dev);
 121 }
 122
 123 static int pc_existing_dimms_capacity_internal(Object *obj, void *opaque)
 124 {
 125     pc_dimms_capacity *cap = opaque;
 126     uint64_t *size = &cap->size;
 127
 128     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 129         DeviceState *dev = DEVICE(obj);
 130
 131         if (dev->realized) {
 132             (*size) += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
 133                 cap->errp);
 134         }
 135
 136         if (cap->errp && *cap->errp) {
 137             return 1;
 138         }
 139     }
 140     object_child_foreach(obj, pc_existing_dimms_capacity_internal, opaque);
 141     return 0;
 142 }
 143
 144 uint64_t pc_existing_dimms_capacity(Error **errp)
 145 {
 146     pc_dimms_capacity cap;
 147
 148     cap.size = 0;
 149     cap.errp = errp;
 150
 151     pc_existing_dimms_capacity_internal(qdev_get_machine(), &cap);
 152     return cap.size;
 153 }
 154
 155 int qmp_pc_dimm_device_list(Object *obj, void *opaque)
 156 {
 157     MemoryDeviceInfoList ***prev = opaque;
 158
 159     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 160         DeviceState *dev = DEVICE(obj);
 161
 162         if (dev->realized) {
 163             MemoryDeviceInfoList *elem = g_new0(MemoryDeviceInfoList, 1);
 164             MemoryDeviceInfo *info = g_new0(MemoryDeviceInfo, 1);
 165             PCDIMMDeviceInfo *di = g_new0(PCDIMMDeviceInfo, 1);
 166             DeviceClass *dc = DEVICE_GET_CLASS(obj);
 167             PCDIMMDevice *dimm = PC_DIMM(obj);
 168
 169             if (dev->id) {
 170                 di->has_id = true;
 171                 di->id = g_strdup(dev->id);
 172             }
 173             di->hotplugged = dev->hotplugged;
 174             di->hotpluggable = dc->hotpluggable;
 175             di->addr = dimm->addr;
 176             di->slot = dimm->slot;
 177             di->node = dimm->node;
 178             di->size = object_property_get_int(OBJECT(dimm), PC_DIMM_SIZE_PROP,
 179                                                NULL);
 180             di->memdev = object_get_canonical_path(OBJECT(dimm->hostmem));
 181
 182             info->u.dimm = di;
 183             elem->value = info;
 184             elem->next = NULL;
 185             **prev = elem;
 186             *prev = &elem->next;
 187         }
 188     }
 189
 190     object_child_foreach(obj, qmp_pc_dimm_device_list, opaque);
 191     return 0;
 192 }
 193
 194 ram_addr_t get_current_ram_size(void)
 195 {
 196     MemoryDeviceInfoList *info_list = NULL;
 197     MemoryDeviceInfoList **prev = &info_list;
 198     MemoryDeviceInfoList *info;
 199     ram_addr_t size = ram_size;
 200
 201     qmp_pc_dimm_device_list(qdev_get_machine(), &prev);
 202     for (info = info_list; info; info = info->next) {
 203         MemoryDeviceInfo *value = info->value;
 204
 205         if (value) {
 206             switch (value->type) {
 207             case MEMORY_DEVICE_INFO_KIND_DIMM:
 208                 size += value->u.dimm->size;
 209                 break;
 210             default:
 211                 break;
 212             }
 213         }
 214     }
 215     qapi_free_MemoryDeviceInfoList(info_list);
 216
 217     return size;
 218 }
 219
 220 static int pc_dimm_slot2bitmap(Object *obj, void *opaque)
 221 {
 222     unsigned long *bitmap = opaque;
 223
 224     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 225         DeviceState *dev = DEVICE(obj);
 226         if (dev->realized) { /* count only realized DIMMs */
 227             PCDIMMDevice *d = PC_DIMM(obj);
 228             set_bit(d->slot, bitmap);
 229         }
 230     }
 231
 232     object_child_foreach(obj, pc_dimm_slot2bitmap, opaque);
 233     return 0;
 234 }
 235
 236 int pc_dimm_get_free_slot(const int *hint, int max_slots, Error **errp)
 237 {
 238     unsigned long *bitmap = bitmap_new(max_slots);
 239     int slot = 0;
 240
 241     object_child_foreach(qdev_get_machine(), pc_dimm_slot2bitmap, bitmap);
 242
 243     /* check if requested slot is not occupied */
 244     if (hint) {
 245         if (*hint >= max_slots) {
 246             error_setg(errp, "invalid slot# %d, should be less than %d",
 247                        *hint, max_slots);
 248         } else if (!test_bit(*hint, bitmap)) {
 249             slot = *hint;
 250         } else {
 251             error_setg(errp, "slot %d is busy", *hint);
 252         }
 253         goto out;
 254     }
 255
 256     /* search for free slot */
 257     slot = find_first_zero_bit(bitmap, max_slots);
 258     if (slot == max_slots) {
 259         error_setg(errp, "no free slots available");
 260     }
 261 out:
 262     g_free(bitmap);
 263     return slot;
 264 }
 265
 266 static gint pc_dimm_addr_sort(gconstpointer a, gconstpointer b)
 267 {
 268     PCDIMMDevice *x = PC_DIMM(a);
 269     PCDIMMDevice *y = PC_DIMM(b);
 270     Int128 diff = int128_sub(int128_make64(x->addr), int128_make64(y->addr));
 271
 272     if (int128_lt(diff, int128_zero())) {
 273         return -1;
 274     } else if (int128_gt(diff, int128_zero())) {
 275         return 1;
 276     }
 277     return 0;
 278 }
 279
 280 static int pc_dimm_built_list(Object *obj, void *opaque)
 281 {
 282     GSList **list = opaque;
 283
 284     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 285         DeviceState *dev = DEVICE(obj);
 286         if (dev->realized) { /* only realized DIMMs matter */
 287             *list = g_slist_insert_sorted(*list, dev, pc_dimm_addr_sort);
 288         }
 289     }
 290
 291     object_child_foreach(obj, pc_dimm_built_list, opaque);
 292     return 0;
 293 }
 294
 295 uint64_t pc_dimm_get_free_addr(uint64_t address_space_start,
 296                                uint64_t address_space_size,
 297                                uint64_t *hint, uint64_t align, uint64_t size,
 298                                Error **errp)
 299 {
 300     GSList *list = NULL, *item;
 301     uint64_t new_addr, ret = 0;
 302     uint64_t address_space_end = address_space_start + address_space_size;
 303
 304     g_assert(QEMU_ALIGN_UP(address_space_start, align) == address_space_start);
 305
 306     if (!address_space_size) {
 307         error_setg(errp, "memory hotplug is not enabled, "
 308                          "please add maxmem option");
 309         goto out;
 310     }
 311
 312     if (hint && QEMU_ALIGN_UP(*hint, align) != *hint) {
 313         error_setg(errp, "address must be aligned to 0x%" PRIx64 " bytes",
 314                    align);
 315         goto out;
 316     }
 317
 318     if (QEMU_ALIGN_UP(size, align) != size) {
 319         error_setg(errp, "backend memory size must be multiple of 0x%"
 320                    PRIx64, align);
 321         goto out;
 322     }
 323
 324     assert(address_space_end > address_space_start);
 325     object_child_foreach(qdev_get_machine(), pc_dimm_built_list, &list);
 326
 327     if (hint) {
 328         new_addr = *hint;
 329     } else {
 330         new_addr = address_space_start;
 331     }
 332
 333     /* find address range that will fit new DIMM */
 334     for (item = list; item; item = g_slist_next(item)) {
 335         PCDIMMDevice *dimm = item->data;
 336         uint64_t dimm_size = object_property_get_int(OBJECT(dimm),
 337                                                      PC_DIMM_SIZE_PROP,
 338                                                      errp);
 339         if (errp && *errp) {
 340             goto out;
 341         }
 342
 343         if (ranges_overlap(dimm->addr, dimm_size, new_addr, size)) {
 344             if (hint) {
 345                 DeviceState *d = DEVICE(dimm);
 346                 error_setg(errp, "address range conflicts with '%s'", d->id);
 347                 goto out;
 348             }
 349             new_addr = QEMU_ALIGN_UP(dimm->addr + dimm_size, align);
 350         }
 351     }
 352     ret = new_addr;
 353
 354     if (new_addr < address_space_start) {
 355         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 356                    "] at 0x%" PRIx64, new_addr, size, address_space_start);
 357     } else if ((new_addr + size) > address_space_end) {
 358         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 359                    "] beyond 0x%" PRIx64, new_addr, size, address_space_end);
 360     }
 361
 362 out:
 363     g_slist_free(list);
 364     return ret;
 365 }
 366
 367 static Property pc_dimm_properties[] = {
 368     DEFINE_PROP_UINT64(PC_DIMM_ADDR_PROP, PCDIMMDevice, addr, 0),
 369     DEFINE_PROP_UINT32(PC_DIMM_NODE_PROP, PCDIMMDevice, node, 0),
 370     DEFINE_PROP_INT32(PC_DIMM_SLOT_PROP, PCDIMMDevice, slot,
 371                       PC_DIMM_UNASSIGNED_SLOT),
 372     DEFINE_PROP_END_OF_LIST(),
 373 };
 374
 375 static void pc_dimm_get_size(Object *obj, Visitor *v, void *opaque,
 376                           const char *name, Error **errp)
 377 {
 378     int64_t value;
 379     MemoryRegion *mr;
 380     PCDIMMDevice *dimm = PC_DIMM(obj);
 381
 382     mr = host_memory_backend_get_memory(dimm->hostmem, errp);
 383     value = memory_region_size(mr);
 384
 385     visit_type_int(v, &value, name, errp);
 386 }
 387
 388 static void pc_dimm_check_memdev_is_busy(Object *obj, const char *name,
 389                                       Object *val, Error **errp)
 390 {
 391     MemoryRegion *mr;
 392
 393     mr = host_memory_backend_get_memory(MEMORY_BACKEND(val), errp);
 394     if (memory_region_is_mapped(mr)) {
 395         char *path = object_get_canonical_path_component(val);
 396         error_setg(errp, "can't use already busy memdev: %s", path);
 397         g_free(path);
 398     } else {
 399         qdev_prop_allow_set_link_before_realize(obj, name, val, errp);
 400     }
 401 }
 402
 403 static void pc_dimm_init(Object *obj)
 404 {
 405     PCDIMMDevice *dimm = PC_DIMM(obj);
 406
 407     object_property_add(obj, PC_DIMM_SIZE_PROP, "int", pc_dimm_get_size,
 408                         NULL, NULL, NULL, &error_abort);
 409     object_property_add_link(obj, PC_DIMM_MEMDEV_PROP, TYPE_MEMORY_BACKEND,
 410                              (Object **)&dimm->hostmem,
 411                              pc_dimm_check_memdev_is_busy,
 412                              OBJ_PROP_LINK_UNREF_ON_RELEASE,
 413                              &error_abort);
 414 }
 415
 416 static void pc_dimm_realize(DeviceState *dev, Error **errp)
 417 {
 418     PCDIMMDevice *dimm = PC_DIMM(dev);
 419
 420     if (!dimm->hostmem) {
 421         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property is not set");
 422         return;
 423     }
 424     if (((nb_numa_nodes > 0) && (dimm->node >= nb_numa_nodes)) ||
 425         (!nb_numa_nodes && dimm->node)) {
 426         error_setg(errp, "'DIMM property " PC_DIMM_NODE_PROP " has value %"
 427                    PRIu32 "' which exceeds the number of numa nodes: %d",
 428                    dimm->node, nb_numa_nodes ? nb_numa_nodes : 1);
 429         return;
 430     }
 431 }
 432
 433 static MemoryRegion *pc_dimm_get_memory_region(PCDIMMDevice *dimm)
 434 {
 435     return host_memory_backend_get_memory(dimm->hostmem, &error_abort);
 436 }
 437
 438 static void pc_dimm_class_init(ObjectClass *oc, void *data)
 439 {
 440     DeviceClass *dc = DEVICE_CLASS(oc);
 441     PCDIMMDeviceClass *ddc = PC_DIMM_CLASS(oc);
 442
 443     dc->realize = pc_dimm_realize;
 444     dc->props = pc_dimm_properties;
 445     dc->desc = "DIMM memory module";
 446
 447     ddc->get_memory_region = pc_dimm_get_memory_region;
 448 }
 449
 450 static TypeInfo pc_dimm_info = {
 451     .name          = TYPE_PC_DIMM,
 452     .parent        = TYPE_DEVICE,
 453     .instance_size = sizeof(PCDIMMDevice),
 454     .instance_init = pc_dimm_init,
 455     .class_init    = pc_dimm_class_init,
 456     .class_size    = sizeof(PCDIMMDeviceClass),
 457 };
 458
 459 static void pc_dimm_register_types(void)
 460 {
 461     type_register_static(&pc_dimm_info);
 462 }
 463
 464 type_init(pc_dimm_register_types)