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