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
 113 out:
 114     error_propagate(errp, local_err);
 115 }
 116
 117 void pc_dimm_memory_unplug(DeviceState *dev, MemoryHotplugState *hpms,
 118                            MemoryRegion *mr)
 119 {
 120     PCDIMMDevice *dimm = PC_DIMM(dev);
 121     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
 122     MemoryRegion *vmstate_mr = ddc->get_vmstate_memory_region(dimm);
 123
 124     memory_region_del_subregion(&hpms->mr, mr);
 125     vmstate_unregister_ram(vmstate_mr, dev);
 126 }
 127
 128 static int pc_existing_dimms_capacity_internal(Object *obj, void *opaque)
 129 {
 130     pc_dimms_capacity *cap = opaque;
 131     uint64_t *size = &cap->size;
 132
 133     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 134         DeviceState *dev = DEVICE(obj);
 135
 136         if (dev->realized) {
 137             (*size) += object_property_get_uint(obj, PC_DIMM_SIZE_PROP,
 138                 cap->errp);
 139         }
 140
 141         if (cap->errp && *cap->errp) {
 142             return 1;
 143         }
 144     }
 145     object_child_foreach(obj, pc_existing_dimms_capacity_internal, opaque);
 146     return 0;
 147 }
 148
 149 uint64_t pc_existing_dimms_capacity(Error **errp)
 150 {
 151     pc_dimms_capacity cap;
 152
 153     cap.size = 0;
 154     cap.errp = errp;
 155
 156     pc_existing_dimms_capacity_internal(qdev_get_machine(), &cap);
 157     return cap.size;
 158 }
 159
 160 uint64_t get_plugged_memory_size(void)
 161 {
 162     return pc_existing_dimms_capacity(&error_abort);
 163 }
 164
 165 int qmp_pc_dimm_device_list(Object *obj, void *opaque)
 166 {
 167     MemoryDeviceInfoList ***prev = opaque;
 168
 169     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 170         DeviceState *dev = DEVICE(obj);
 171
 172         if (dev->realized) {
 173             MemoryDeviceInfoList *elem = g_new0(MemoryDeviceInfoList, 1);
 174             MemoryDeviceInfo *info = g_new0(MemoryDeviceInfo, 1);
 175             PCDIMMDeviceInfo *di = g_new0(PCDIMMDeviceInfo, 1);
 176             DeviceClass *dc = DEVICE_GET_CLASS(obj);
 177             PCDIMMDevice *dimm = PC_DIMM(obj);
 178
 179             if (dev->id) {
 180                 di->has_id = true;
 181                 di->id = g_strdup(dev->id);
 182             }
 183             di->hotplugged = dev->hotplugged;
 184             di->hotpluggable = dc->hotpluggable;
 185             di->addr = dimm->addr;
 186             di->slot = dimm->slot;
 187             di->node = dimm->node;
 188             di->size = object_property_get_uint(OBJECT(dimm), PC_DIMM_SIZE_PROP,
 189                                                 NULL);
 190             di->memdev = object_get_canonical_path(OBJECT(dimm->hostmem));
 191
 192             info->u.dimm.data = di;
 193             elem->value = info;
 194             elem->next = NULL;
 195             **prev = elem;
 196             *prev = &elem->next;
 197         }
 198     }
 199
 200     object_child_foreach(obj, qmp_pc_dimm_device_list, opaque);
 201     return 0;
 202 }
 203
 204 static int pc_dimm_slot2bitmap(Object *obj, void *opaque)
 205 {
 206     unsigned long *bitmap = opaque;
 207
 208     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 209         DeviceState *dev = DEVICE(obj);
 210         if (dev->realized) { /* count only realized DIMMs */
 211             PCDIMMDevice *d = PC_DIMM(obj);
 212             set_bit(d->slot, bitmap);
 213         }
 214     }
 215
 216     object_child_foreach(obj, pc_dimm_slot2bitmap, opaque);
 217     return 0;
 218 }
 219
 220 int pc_dimm_get_free_slot(const int *hint, int max_slots, Error **errp)
 221 {
 222     unsigned long *bitmap = bitmap_new(max_slots);
 223     int slot = 0;
 224
 225     object_child_foreach(qdev_get_machine(), pc_dimm_slot2bitmap, bitmap);
 226
 227     /* check if requested slot is not occupied */
 228     if (hint) {
 229         if (*hint >= max_slots) {
 230             error_setg(errp, "invalid slot# %d, should be less than %d",
 231                        *hint, max_slots);
 232         } else if (!test_bit(*hint, bitmap)) {
 233             slot = *hint;
 234         } else {
 235             error_setg(errp, "slot %d is busy", *hint);
 236         }
 237         goto out;
 238     }
 239
 240     /* search for free slot */
 241     slot = find_first_zero_bit(bitmap, max_slots);
 242     if (slot == max_slots) {
 243         error_setg(errp, "no free slots available");
 244     }
 245 out:
 246     g_free(bitmap);
 247     return slot;
 248 }
 249
 250 static gint pc_dimm_addr_sort(gconstpointer a, gconstpointer b)
 251 {
 252     PCDIMMDevice *x = PC_DIMM(a);
 253     PCDIMMDevice *y = PC_DIMM(b);
 254     Int128 diff = int128_sub(int128_make64(x->addr), int128_make64(y->addr));
 255
 256     if (int128_lt(diff, int128_zero())) {
 257         return -1;
 258     } else if (int128_gt(diff, int128_zero())) {
 259         return 1;
 260     }
 261     return 0;
 262 }
 263
 264 static int pc_dimm_built_list(Object *obj, void *opaque)
 265 {
 266     GSList **list = opaque;
 267
 268     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
 269         DeviceState *dev = DEVICE(obj);
 270         if (dev->realized) { /* only realized DIMMs matter */
 271             *list = g_slist_insert_sorted(*list, dev, pc_dimm_addr_sort);
 272         }
 273     }
 274
 275     object_child_foreach(obj, pc_dimm_built_list, opaque);
 276     return 0;
 277 }
 278
 279 uint64_t pc_dimm_get_free_addr(uint64_t address_space_start,
 280                                uint64_t address_space_size,
 281                                uint64_t *hint, uint64_t align, uint64_t size,
 282                                Error **errp)
 283 {
 284     GSList *list = NULL, *item;
 285     uint64_t new_addr, ret = 0;
 286     uint64_t address_space_end = address_space_start + address_space_size;
 287
 288     g_assert(QEMU_ALIGN_UP(address_space_start, align) == address_space_start);
 289
 290     if (!address_space_size) {
 291         error_setg(errp, "memory hotplug is not enabled, "
 292                          "please add maxmem option");
 293         goto out;
 294     }
 295
 296     if (hint && QEMU_ALIGN_UP(*hint, align) != *hint) {
 297         error_setg(errp, "address must be aligned to 0x%" PRIx64 " bytes",
 298                    align);
 299         goto out;
 300     }
 301
 302     if (QEMU_ALIGN_UP(size, align) != size) {
 303         error_setg(errp, "backend memory size must be multiple of 0x%"
 304                    PRIx64, align);
 305         goto out;
 306     }
 307
 308     assert(address_space_end > address_space_start);
 309     object_child_foreach(qdev_get_machine(), pc_dimm_built_list, &list);
 310
 311     if (hint) {
 312         new_addr = *hint;
 313     } else {
 314         new_addr = address_space_start;
 315     }
 316
 317     /* find address range that will fit new DIMM */
 318     for (item = list; item; item = g_slist_next(item)) {
 319         PCDIMMDevice *dimm = item->data;
 320         uint64_t dimm_size = object_property_get_uint(OBJECT(dimm),
 321                                                       PC_DIMM_SIZE_PROP,
 322                                                       errp);
 323         if (errp && *errp) {
 324             goto out;
 325         }
 326
 327         if (ranges_overlap(dimm->addr, dimm_size, new_addr, size)) {
 328             if (hint) {
 329                 DeviceState *d = DEVICE(dimm);
 330                 error_setg(errp, "address range conflicts with '%s'", d->id);
 331                 goto out;
 332             }
 333             new_addr = QEMU_ALIGN_UP(dimm->addr + dimm_size, align);
 334         }
 335     }
 336     ret = new_addr;
 337
 338     if (new_addr < address_space_start) {
 339         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 340                    "] at 0x%" PRIx64, new_addr, size, address_space_start);
 341     } else if ((new_addr + size) > address_space_end) {
 342         error_setg(errp, "can't add memory [0x%" PRIx64 ":0x%" PRIx64
 343                    "] beyond 0x%" PRIx64, new_addr, size, address_space_end);
 344     }
 345
 346 out:
 347     g_slist_free(list);
 348     return ret;
 349 }
 350
 351 static Property pc_dimm_properties[] = {
 352     DEFINE_PROP_UINT64(PC_DIMM_ADDR_PROP, PCDIMMDevice, addr, 0),
 353     DEFINE_PROP_UINT32(PC_DIMM_NODE_PROP, PCDIMMDevice, node, 0),
 354     DEFINE_PROP_INT32(PC_DIMM_SLOT_PROP, PCDIMMDevice, slot,
 355                       PC_DIMM_UNASSIGNED_SLOT),
 356     DEFINE_PROP_LINK(PC_DIMM_MEMDEV_PROP, PCDIMMDevice, hostmem,
 357                      TYPE_MEMORY_BACKEND, HostMemoryBackend *),
 358     DEFINE_PROP_END_OF_LIST(),
 359 };
 360
 361 static void pc_dimm_get_size(Object *obj, Visitor *v, const char *name,
 362                              void *opaque, Error **errp)
 363 {
 364     uint64_t value;
 365     MemoryRegion *mr;
 366     PCDIMMDevice *dimm = PC_DIMM(obj);
 367     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(obj);
 368
 369     mr = ddc->get_memory_region(dimm, errp);
 370     if (!mr) {
 371         return;
 372     }
 373     value = memory_region_size(mr);
 374
 375     visit_type_uint64(v, name, &value, errp);
 376 }
 377
 378 static void pc_dimm_init(Object *obj)
 379 {
 380     object_property_add(obj, PC_DIMM_SIZE_PROP, "uint64", pc_dimm_get_size,
 381                         NULL, NULL, NULL, &error_abort);
 382 }
 383
 384 static void pc_dimm_realize(DeviceState *dev, Error **errp)
 385 {
 386     PCDIMMDevice *dimm = PC_DIMM(dev);
 387     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
 388
 389     if (!dimm->hostmem) {
 390         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property is not set");
 391         return;
 392     } else if (host_memory_backend_is_mapped(dimm->hostmem)) {
 393         char *path = object_get_canonical_path_component(OBJECT(dimm->hostmem));
 394         error_setg(errp, "can't use already busy memdev: %s", path);
 395         g_free(path);
 396         return;
 397     }
 398     if (((nb_numa_nodes > 0) && (dimm->node >= nb_numa_nodes)) ||
 399         (!nb_numa_nodes && dimm->node)) {
 400         error_setg(errp, "'DIMM property " PC_DIMM_NODE_PROP " has value %"
 401                    PRIu32 "' which exceeds the number of numa nodes: %d",
 402                    dimm->node, nb_numa_nodes ? nb_numa_nodes : 1);
 403         return;
 404     }
 405
 406     if (ddc->realize) {
 407         ddc->realize(dimm, errp);
 408     }
 409
 410     host_memory_backend_set_mapped(dimm->hostmem, true);
 411 }
 412
 413 static void pc_dimm_unrealize(DeviceState *dev, Error **errp)
 414 {
 415     PCDIMMDevice *dimm = PC_DIMM(dev);
 416
 417     host_memory_backend_set_mapped(dimm->hostmem, false);
 418 }
 419
 420 static MemoryRegion *pc_dimm_get_memory_region(PCDIMMDevice *dimm, Error **errp)
 421 {
 422     if (!dimm->hostmem) {
 423         error_setg(errp, "'" PC_DIMM_MEMDEV_PROP "' property must be set");
 424         return NULL;
 425     }
 426
 427     return host_memory_backend_get_memory(dimm->hostmem, errp);
 428 }
 429
 430 static MemoryRegion *pc_dimm_get_vmstate_memory_region(PCDIMMDevice *dimm)
 431 {
 432     return host_memory_backend_get_memory(dimm->hostmem, &error_abort);
 433 }
 434
 435 static void pc_dimm_class_init(ObjectClass *oc, void *data)
 436 {
 437     DeviceClass *dc = DEVICE_CLASS(oc);
 438     PCDIMMDeviceClass *ddc = PC_DIMM_CLASS(oc);
 439
 440     dc->realize = pc_dimm_realize;
 441     dc->unrealize = pc_dimm_unrealize;
 442     dc->props = pc_dimm_properties;
 443     dc->desc = "DIMM memory module";
 444
 445     ddc->get_memory_region = pc_dimm_get_memory_region;
 446     ddc->get_vmstate_memory_region = pc_dimm_get_vmstate_memory_region;
 447 }
 448
 449 static TypeInfo pc_dimm_info = {
 450     .name          = TYPE_PC_DIMM,
 451     .parent        = TYPE_DEVICE,
 452     .instance_size = sizeof(PCDIMMDevice),
 453     .instance_init = pc_dimm_init,
 454     .class_init    = pc_dimm_class_init,
 455     .class_size    = sizeof(PCDIMMDeviceClass),
 456 };
 457
 458 static void pc_dimm_register_types(void)
 459 {
 460     type_register_static(&pc_dimm_info);
 461 }
 462
 463 type_init(pc_dimm_register_types)