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 uint64_t get_plugged_memory_size(void)
 163 {
 164     return pc_existing_dimms_capacity(&error_abort);
 165 }
 166
 167 int qmp_pc_dimm_device_list(Object *obj, void *opaque)
 168 {
 169     MemoryDeviceInfoList ***prev = opaque;
 170
 171     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 172         DeviceState *dev = DEVICE(obj);
 173
 174         if (dev->realized) {
 175             MemoryDeviceInfoList *elem = g_new0(MemoryDeviceInfoList, 1);
 176             MemoryDeviceInfo *info = g_new0(MemoryDeviceInfo, 1);
 177             PCDIMMDeviceInfo *di = g_new0(PCDIMMDeviceInfo, 1);
 178             DeviceClass *dc = DEVICE_GET_CLASS(obj);
 179             PCDIMMDevice *dimm = PC_DIMM(obj);
 180
 181             if (dev->id) {
 182                 di->has_id = true;
 183                 di->id = g_strdup(dev->id);
 184             }
 185             di->hotplugged = dev->hotplugged;
 186             di->hotpluggable = dc->hotpluggable;
 187             di->addr = dimm->addr;
 188             di->slot = dimm->slot;
 189             di->node = dimm->node;
 190             di->size = object_property_get_uint(OBJECT(dimm), PC_DIMM_SIZE_PROP,
 191                                                 NULL);
 192             di->memdev = object_get_canonical_path(OBJECT(dimm->hostmem));
 193
 194             info->u.dimm.data = di;
 195             elem->value = info;
 196             elem->next = NULL;
 197             **prev = elem;
 198             *prev = &elem->next;
 199         }
 200     }
 201
 202     object_child_foreach(obj, qmp_pc_dimm_device_list, opaque);
 203     return 0;
 204 }
 205
 206 static int pc_dimm_slot2bitmap(Object *obj, void *opaque)
 207 {
 208     unsigned long *bitmap = opaque;
 209
 210     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 211         DeviceState *dev = DEVICE(obj);
 212         if (dev->realized) { /* count only realized DIMMs */
 213             PCDIMMDevice *d = PC_DIMM(obj);
 214             set_bit(d->slot, bitmap);
 215         }
 216     }
 217
 218     object_child_foreach(obj, pc_dimm_slot2bitmap, opaque);
 219     return 0;
 220 }
 221
 222 int pc_dimm_get_free_slot(const int *hint, int max_slots, Error **errp)
 223 {
 224     unsigned long *bitmap = bitmap_new(max_slots);
 225     int slot = 0;
 226
 227     object_child_foreach(qdev_get_machine(), pc_dimm_slot2bitmap, bitmap);
 228
 229     /* check if requested slot is not occupied */
 230     if (hint) {
 231         if (*hint >= max_slots) {
 232             error_setg(errp, "invalid slot# %d, should be less than %d",
 233                        *hint, max_slots);
 234         } else if (!test_bit(*hint, bitmap)) {
 235             slot = *hint;
 236         } else {
 237             error_setg(errp, "slot %d is busy", *hint);
 238         }
 239         goto out;
 240     }
 241
 242     /* search for free slot */
 243     slot = find_first_zero_bit(bitmap, max_slots);
 244     if (slot == max_slots) {
 245         error_setg(errp, "no free slots available");
 246     }
 247 out:
 248     g_free(bitmap);
 249     return slot;
 250 }
 251
 252 static gint pc_dimm_addr_sort(gconstpointer a, gconstpointer b)
 253 {
 254     PCDIMMDevice *x = PC_DIMM(a);
 255     PCDIMMDevice *y = PC_DIMM(b);
 256     Int128 diff = int128_sub(int128_make64(x->addr), int128_make64(y->addr));
 257
 258     if (int128_lt(diff, int128_zero())) {
 259         return -1;
 260     } else if (int128_gt(diff, int128_zero())) {
 261         return 1;
 262     }
 263     return 0;
 264 }
 265
 266 static int pc_dimm_built_list(Object *obj, void *opaque)
 267 {
 268     GSList **list = opaque;
 269
 270     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 271         DeviceState *dev = DEVICE(obj);
 272         if (dev->realized) { /* only realized DIMMs matter */
 273             *list = g_slist_insert_sorted(*list, dev, pc_dimm_addr_sort);
 274         }
 275     }
 276
 277     object_child_foreach(obj, pc_dimm_built_list, opaque);
 278     return 0;
 279 }
 280
 281 uint64_t pc_dimm_get_free_addr(uint64_t address_space_start,
 282                                uint64_t address_space_size,
 283                                uint64_t *hint, uint64_t align, uint64_t size,
 284                                Error **errp)
 285 {
 286     GSList *list = NULL, *item;
 287     uint64_t new_addr, ret = 0;
 288     uint64_t address_space_end = address_space_start + address_space_size;
 289
 290     g_assert(QEMU_ALIGN_UP(address_space_start, align) == address_space_start);
 291
 292     if (!address_space_size) {
 293         error_setg(errp, "memory hotplug is not enabled, "
 294                          "please add maxmem option");
 295         goto out;
 296     }
 297
 298     if (hint && QEMU_ALIGN_UP(*hint, align) != *hint) {
 299         error_setg(errp, "address must be aligned to 0x%" PRIx64 " bytes",
 300                    align);
 301         goto out;
 302     }
 303
 304     if (QEMU_ALIGN_UP(size, align) != size) {
 305         error_setg(errp, "backend memory size must be multiple of 0x%"
 306                    PRIx64, align);
 307         goto out;
 308     }
 309
 310     assert(address_space_end > address_space_start);
 311     object_child_foreach(qdev_get_machine(), pc_dimm_built_list, &list);
 312
 313     if (hint) {
 314         new_addr = *hint;
 315     } else {
 316         new_addr = address_space_start;
 317     }
 318
 319     /* find address range that will fit new DIMM */
 320     for (item = list; item; item = g_slist_next(item)) {
 321         PCDIMMDevice *dimm = item->data;
 322         uint64_t dimm_size = object_property_get_uint(OBJECT(dimm),
 323                                                       PC_DIMM_SIZE_PROP,
 324                                                       errp);
 325         if (errp && *errp) {
 326             goto out;
 327         }
 328
 329         if (ranges_overlap(dimm->addr, dimm_size, new_addr, size)) {
 330             if (hint) {
 331                 DeviceState *d = DEVICE(dimm);
 332                 error_setg(errp, "address range conflicts with '%s'", d->id);
 333                 goto out;
 334             }
 335             new_addr = QEMU_ALIGN_UP(dimm->addr + dimm_size, align);
 336         }
 337     }
 338     ret = new_addr;
 339
 340     if (new_addr < address_space_start) {
 341         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 342                    "] at 0x%" PRIx64, new_addr, size, address_space_start);
 343     } else if ((new_addr + size) > address_space_end) {
 344         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 345                    "] beyond 0x%" PRIx64, new_addr, size, address_space_end);
 346     }
 347
 348 out:
 349     g_slist_free(list);
 350     return ret;
 351 }
 352
 353 static Property pc_dimm_properties[] = {
 354     DEFINE_PROP_UINT64(PC_DIMM_ADDR_PROP, PCDIMMDevice, addr, 0),
 355     DEFINE_PROP_UINT32(PC_DIMM_NODE_PROP, PCDIMMDevice, node, 0),
 356     DEFINE_PROP_INT32(PC_DIMM_SLOT_PROP, PCDIMMDevice, slot,
 357                       PC_DIMM_UNASSIGNED_SLOT),
 358     DEFINE_PROP_LINK(PC_DIMM_MEMDEV_PROP, PCDIMMDevice, hostmem,
 359                      TYPE_MEMORY_BACKEND, HostMemoryBackend *),
 360     DEFINE_PROP_END_OF_LIST(),
 361 };
 362
 363 static void pc_dimm_get_size(Object *obj, Visitor *v, const char *name,
 364                              void *opaque, Error **errp)
 365 {
 366     uint64_t value;
 367     MemoryRegion *mr;
 368     PCDIMMDevice *dimm = PC_DIMM(obj);
 369     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(obj);
 370
 371     mr = ddc->get_memory_region(dimm, errp);
 372     if (!mr) {
 373         return;
 374     }
 375     value = memory_region_size(mr);
 376
 377     visit_type_uint64(v, name, &value, errp);
 378 }
 379
 380 static void pc_dimm_init(Object *obj)
 381 {
 382     object_property_add(obj, PC_DIMM_SIZE_PROP, "uint64", pc_dimm_get_size,
 383                         NULL, NULL, NULL, &error_abort);
 384 }
 385
 386 static void pc_dimm_realize(DeviceState *dev, Error **errp)
 387 {
 388     PCDIMMDevice *dimm = PC_DIMM(dev);
 389     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
 390
 391     if (!dimm->hostmem) {
 392         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property is not set");
 393         return;
 394     } else if (host_memory_backend_is_mapped(dimm->hostmem)) {
 395         char *path = object_get_canonical_path_component(OBJECT(dimm->hostmem));
 396         error_setg(errp, "can't use already busy memdev: %s", path);
 397         g_free(path);
 398         return;
 399     }
 400     if (((nb_numa_nodes > 0) && (dimm->node >= nb_numa_nodes)) ||
 401         (!nb_numa_nodes && dimm->node)) {
 402         error_setg(errp, "'DIMM property " PC_DIMM_NODE_PROP " has value %"
 403                    PRIu32 "' which exceeds the number of numa nodes: %d",
 404                    dimm->node, nb_numa_nodes ? nb_numa_nodes : 1);
 405         return;
 406     }
 407
 408     if (ddc->realize) {
 409         ddc->realize(dimm, errp);
 410     }
 411
 412     host_memory_backend_set_mapped(dimm->hostmem, true);
 413 }
 414
 415 static void pc_dimm_unrealize(DeviceState *dev, Error **errp)
 416 {
 417     PCDIMMDevice *dimm = PC_DIMM(dev);
 418
 419     host_memory_backend_set_mapped(dimm->hostmem, false);
 420 }
 421
 422 static MemoryRegion *pc_dimm_get_memory_region(PCDIMMDevice *dimm, Error **errp)
 423 {
 424     if (!dimm->hostmem) {
 425         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property must be set");
 426         return NULL;
 427     }
 428
 429     return host_memory_backend_get_memory(dimm->hostmem, errp);
 430 }
 431
 432 static MemoryRegion *pc_dimm_get_vmstate_memory_region(PCDIMMDevice *dimm)
 433 {
 434     return host_memory_backend_get_memory(dimm->hostmem, &error_abort);
 435 }
 436
 437 static void pc_dimm_class_init(ObjectClass *oc, void *data)
 438 {
 439     DeviceClass *dc = DEVICE_CLASS(oc);
 440     PCDIMMDeviceClass *ddc = PC_DIMM_CLASS(oc);
 441
 442     dc->realize = pc_dimm_realize;
 443     dc->unrealize = pc_dimm_unrealize;
 444     dc->props = pc_dimm_properties;
 445     dc->desc = "DIMM memory module";
 446
 447     ddc->get_memory_region = pc_dimm_get_memory_region;
 448     ddc->get_vmstate_memory_region = pc_dimm_get_vmstate_memory_region;
 449 }
 450
 451 static TypeInfo pc_dimm_info = {
 452     .name          = TYPE_PC_DIMM,
 453     .parent        = TYPE_DEVICE,
 454     .instance_size = sizeof(PCDIMMDevice),
 455     .instance_init = pc_dimm_init,
 456     .class_init    = pc_dimm_class_init,
 457     .class_size    = sizeof(PCDIMMDeviceClass),
 458 };
 459
 460 static void pc_dimm_register_types(void)
 461 {
 462     type_register_static(&pc_dimm_info);
 463 }
 464
 465 type_init(pc_dimm_register_types)