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