1 /* 2 * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator 3 * 4 * Copyright (c) 2004-2007 Fabrice Bellard 5 * Copyright (c) 2007 Jocelyn Mayer 6 * Copyright (c) 2010 David Gibson, IBM Corporation. 7 * 8 * Permission is hereby granted, free of charge, to any person obtaining a copy 9 * of this software and associated documentation files (the "Software"), to deal 10 * in the Software without restriction, including without limitation the rights 11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 12 * copies of the Software, and to permit persons to whom the Software is 13 * furnished to do so, subject to the following conditions: 14 * 15 * The above copyright notice and this permission notice shall be included in 16 * all copies or substantial portions of the Software. 17 * 18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 24 * THE SOFTWARE. 25 */ 26 27 #include "qemu/osdep.h" 28 #include "qemu-common.h" 29 #include "qemu/datadir.h" 30 #include "qapi/error.h" 31 #include "qapi/qapi-events-machine.h" 32 #include "qapi/visitor.h" 33 #include "sysemu/sysemu.h" 34 #include "sysemu/hostmem.h" 35 #include "sysemu/numa.h" 36 #include "sysemu/qtest.h" 37 #include "sysemu/reset.h" 38 #include "sysemu/runstate.h" 39 #include "qemu/log.h" 40 #include "hw/fw-path-provider.h" 41 #include "elf.h" 42 #include "net/net.h" 43 #include "sysemu/device_tree.h" 44 #include "sysemu/cpus.h" 45 #include "sysemu/hw_accel.h" 46 #include "kvm_ppc.h" 47 #include "migration/misc.h" 48 #include "migration/qemu-file-types.h" 49 #include "migration/global_state.h" 50 #include "migration/register.h" 51 #include "migration/blocker.h" 52 #include "mmu-hash64.h" 53 #include "mmu-book3s-v3.h" 54 #include "cpu-models.h" 55 #include "hw/core/cpu.h" 56 57 #include "hw/ppc/ppc.h" 58 #include "hw/loader.h" 59 60 #include "hw/ppc/fdt.h" 61 #include "hw/ppc/spapr.h" 62 #include "hw/ppc/spapr_vio.h" 63 #include "hw/qdev-properties.h" 64 #include "hw/pci-host/spapr.h" 65 #include "hw/pci/msi.h" 66 67 #include "hw/pci/pci.h" 68 #include "hw/scsi/scsi.h" 69 #include "hw/virtio/virtio-scsi.h" 70 #include "hw/virtio/vhost-scsi-common.h" 71 72 #include "exec/ram_addr.h" 73 #include "hw/usb.h" 74 #include "qemu/config-file.h" 75 #include "qemu/error-report.h" 76 #include "trace.h" 77 #include "hw/nmi.h" 78 #include "hw/intc/intc.h" 79 80 #include "hw/ppc/spapr_cpu_core.h" 81 #include "hw/mem/memory-device.h" 82 #include "hw/ppc/spapr_tpm_proxy.h" 83 #include "hw/ppc/spapr_nvdimm.h" 84 #include "hw/ppc/spapr_numa.h" 85 #include "hw/ppc/pef.h" 86 87 #include "monitor/monitor.h" 88 89 #include <libfdt.h> 90 91 /* SLOF memory layout: 92 * 93 * SLOF raw image loaded at 0, copies its romfs right below the flat 94 * device-tree, then position SLOF itself 31M below that 95 * 96 * So we set FW_OVERHEAD to 40MB which should account for all of that 97 * and more 98 * 99 * We load our kernel at 4M, leaving space for SLOF initial image 100 */ 101 #define FDT_MAX_ADDR 0x80000000 /* FDT must stay below that */ 102 #define FW_MAX_SIZE 0x400000 103 #define FW_FILE_NAME "slof.bin" 104 #define FW_FILE_NAME_VOF "vof.bin" 105 #define FW_OVERHEAD 0x2800000 106 #define KERNEL_LOAD_ADDR FW_MAX_SIZE 107 108 #define MIN_RMA_SLOF (128 * MiB) 109 110 #define PHANDLE_INTC 0x00001111 111 112 /* These two functions implement the VCPU id numbering: one to compute them 113 * all and one to identify thread 0 of a VCORE. Any change to the first one 114 * is likely to have an impact on the second one, so let's keep them close. 115 */ 116 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index) 117 { 118 MachineState *ms = MACHINE(spapr); 119 unsigned int smp_threads = ms->smp.threads; 120 121 assert(spapr->vsmt); 122 return 123 (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads; 124 } 125 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr, 126 PowerPCCPU *cpu) 127 { 128 assert(spapr->vsmt); 129 return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0; 130 } 131 132 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque) 133 { 134 /* Dummy entries correspond to unused ICPState objects in older QEMUs, 135 * and newer QEMUs don't even have them. In both cases, we don't want 136 * to send anything on the wire. 137 */ 138 return false; 139 } 140 141 static const VMStateDescription pre_2_10_vmstate_dummy_icp = { 142 .name = "icp/server", 143 .version_id = 1, 144 .minimum_version_id = 1, 145 .needed = pre_2_10_vmstate_dummy_icp_needed, 146 .fields = (VMStateField[]) { 147 VMSTATE_UNUSED(4), /* uint32_t xirr */ 148 VMSTATE_UNUSED(1), /* uint8_t pending_priority */ 149 VMSTATE_UNUSED(1), /* uint8_t mfrr */ 150 VMSTATE_END_OF_LIST() 151 }, 152 }; 153 154 static void pre_2_10_vmstate_register_dummy_icp(int i) 155 { 156 vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp, 157 (void *)(uintptr_t) i); 158 } 159 160 static void pre_2_10_vmstate_unregister_dummy_icp(int i) 161 { 162 vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp, 163 (void *)(uintptr_t) i); 164 } 165 166 int spapr_max_server_number(SpaprMachineState *spapr) 167 { 168 MachineState *ms = MACHINE(spapr); 169 170 assert(spapr->vsmt); 171 return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads); 172 } 173 174 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu, 175 int smt_threads) 176 { 177 int i, ret = 0; 178 uint32_t servers_prop[smt_threads]; 179 uint32_t gservers_prop[smt_threads * 2]; 180 int index = spapr_get_vcpu_id(cpu); 181 182 if (cpu->compat_pvr) { 183 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr); 184 if (ret < 0) { 185 return ret; 186 } 187 } 188 189 /* Build interrupt servers and gservers properties */ 190 for (i = 0; i < smt_threads; i++) { 191 servers_prop[i] = cpu_to_be32(index + i); 192 /* Hack, direct the group queues back to cpu 0 */ 193 gservers_prop[i*2] = cpu_to_be32(index + i); 194 gservers_prop[i*2 + 1] = 0; 195 } 196 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s", 197 servers_prop, sizeof(servers_prop)); 198 if (ret < 0) { 199 return ret; 200 } 201 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s", 202 gservers_prop, sizeof(gservers_prop)); 203 204 return ret; 205 } 206 207 static void spapr_dt_pa_features(SpaprMachineState *spapr, 208 PowerPCCPU *cpu, 209 void *fdt, int offset) 210 { 211 uint8_t pa_features_206[] = { 6, 0, 212 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 }; 213 uint8_t pa_features_207[] = { 24, 0, 214 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, 215 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 216 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, 217 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 }; 218 uint8_t pa_features_300[] = { 66, 0, 219 /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */ 220 /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */ 221 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */ 222 /* 6: DS207 */ 223 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */ 224 /* 16: Vector */ 225 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */ 226 /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */ 227 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */ 228 /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */ 229 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */ 230 /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */ 231 0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */ 232 /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */ 233 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */ 234 /* 42: PM, 44: PC RA, 46: SC vec'd */ 235 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */ 236 /* 48: SIMD, 50: QP BFP, 52: String */ 237 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */ 238 /* 54: DecFP, 56: DecI, 58: SHA */ 239 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */ 240 /* 60: NM atomic, 62: RNG */ 241 0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */ 242 }; 243 uint8_t *pa_features = NULL; 244 size_t pa_size; 245 246 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) { 247 pa_features = pa_features_206; 248 pa_size = sizeof(pa_features_206); 249 } 250 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) { 251 pa_features = pa_features_207; 252 pa_size = sizeof(pa_features_207); 253 } 254 if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) { 255 pa_features = pa_features_300; 256 pa_size = sizeof(pa_features_300); 257 } 258 if (!pa_features) { 259 return; 260 } 261 262 if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) { 263 /* 264 * Note: we keep CI large pages off by default because a 64K capable 265 * guest provisioned with large pages might otherwise try to map a qemu 266 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages 267 * even if that qemu runs on a 4k host. 268 * We dd this bit back here if we are confident this is not an issue 269 */ 270 pa_features[3] |= 0x20; 271 } 272 if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) { 273 pa_features[24] |= 0x80; /* Transactional memory support */ 274 } 275 if (spapr->cas_pre_isa3_guest && pa_size > 40) { 276 /* Workaround for broken kernels that attempt (guest) radix 277 * mode when they can't handle it, if they see the radix bit set 278 * in pa-features. So hide it from them. */ 279 pa_features[40 + 2] &= ~0x80; /* Radix MMU */ 280 } 281 282 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size))); 283 } 284 285 static hwaddr spapr_node0_size(MachineState *machine) 286 { 287 if (machine->numa_state->num_nodes) { 288 int i; 289 for (i = 0; i < machine->numa_state->num_nodes; ++i) { 290 if (machine->numa_state->nodes[i].node_mem) { 291 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem), 292 machine->ram_size); 293 } 294 } 295 } 296 return machine->ram_size; 297 } 298 299 static void add_str(GString *s, const gchar *s1) 300 { 301 g_string_append_len(s, s1, strlen(s1) + 1); 302 } 303 304 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid, 305 hwaddr start, hwaddr size) 306 { 307 char mem_name[32]; 308 uint64_t mem_reg_property[2]; 309 int off; 310 311 mem_reg_property[0] = cpu_to_be64(start); 312 mem_reg_property[1] = cpu_to_be64(size); 313 314 sprintf(mem_name, "memory@%" HWADDR_PRIx, start); 315 off = fdt_add_subnode(fdt, 0, mem_name); 316 _FDT(off); 317 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory"))); 318 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property, 319 sizeof(mem_reg_property)))); 320 spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid); 321 return off; 322 } 323 324 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr) 325 { 326 MemoryDeviceInfoList *info; 327 328 for (info = list; info; info = info->next) { 329 MemoryDeviceInfo *value = info->value; 330 331 if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) { 332 PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data; 333 334 if (addr >= pcdimm_info->addr && 335 addr < (pcdimm_info->addr + pcdimm_info->size)) { 336 return pcdimm_info->node; 337 } 338 } 339 } 340 341 return -1; 342 } 343 344 struct sPAPRDrconfCellV2 { 345 uint32_t seq_lmbs; 346 uint64_t base_addr; 347 uint32_t drc_index; 348 uint32_t aa_index; 349 uint32_t flags; 350 } QEMU_PACKED; 351 352 typedef struct DrconfCellQueue { 353 struct sPAPRDrconfCellV2 cell; 354 QSIMPLEQ_ENTRY(DrconfCellQueue) entry; 355 } DrconfCellQueue; 356 357 static DrconfCellQueue * 358 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr, 359 uint32_t drc_index, uint32_t aa_index, 360 uint32_t flags) 361 { 362 DrconfCellQueue *elem; 363 364 elem = g_malloc0(sizeof(*elem)); 365 elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs); 366 elem->cell.base_addr = cpu_to_be64(base_addr); 367 elem->cell.drc_index = cpu_to_be32(drc_index); 368 elem->cell.aa_index = cpu_to_be32(aa_index); 369 elem->cell.flags = cpu_to_be32(flags); 370 371 return elem; 372 } 373 374 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt, 375 int offset, MemoryDeviceInfoList *dimms) 376 { 377 MachineState *machine = MACHINE(spapr); 378 uint8_t *int_buf, *cur_index; 379 int ret; 380 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE; 381 uint64_t addr, cur_addr, size; 382 uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size); 383 uint64_t mem_end = machine->device_memory->base + 384 memory_region_size(&machine->device_memory->mr); 385 uint32_t node, buf_len, nr_entries = 0; 386 SpaprDrc *drc; 387 DrconfCellQueue *elem, *next; 388 MemoryDeviceInfoList *info; 389 QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue 390 = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue); 391 392 /* Entry to cover RAM and the gap area */ 393 elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1, 394 SPAPR_LMB_FLAGS_RESERVED | 395 SPAPR_LMB_FLAGS_DRC_INVALID); 396 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry); 397 nr_entries++; 398 399 cur_addr = machine->device_memory->base; 400 for (info = dimms; info; info = info->next) { 401 PCDIMMDeviceInfo *di = info->value->u.dimm.data; 402 403 addr = di->addr; 404 size = di->size; 405 node = di->node; 406 407 /* 408 * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The 409 * area is marked hotpluggable in the next iteration for the bigger 410 * chunk including the NVDIMM occupied area. 411 */ 412 if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM) 413 continue; 414 415 /* Entry for hot-pluggable area */ 416 if (cur_addr < addr) { 417 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size); 418 g_assert(drc); 419 elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size, 420 cur_addr, spapr_drc_index(drc), -1, 0); 421 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry); 422 nr_entries++; 423 } 424 425 /* Entry for DIMM */ 426 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size); 427 g_assert(drc); 428 elem = spapr_get_drconf_cell(size / lmb_size, addr, 429 spapr_drc_index(drc), node, 430 (SPAPR_LMB_FLAGS_ASSIGNED | 431 SPAPR_LMB_FLAGS_HOTREMOVABLE)); 432 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry); 433 nr_entries++; 434 cur_addr = addr + size; 435 } 436 437 /* Entry for remaining hotpluggable area */ 438 if (cur_addr < mem_end) { 439 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size); 440 g_assert(drc); 441 elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size, 442 cur_addr, spapr_drc_index(drc), -1, 0); 443 QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry); 444 nr_entries++; 445 } 446 447 buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t); 448 int_buf = cur_index = g_malloc0(buf_len); 449 *(uint32_t *)int_buf = cpu_to_be32(nr_entries); 450 cur_index += sizeof(nr_entries); 451 452 QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) { 453 memcpy(cur_index, &elem->cell, sizeof(elem->cell)); 454 cur_index += sizeof(elem->cell); 455 QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry); 456 g_free(elem); 457 } 458 459 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len); 460 g_free(int_buf); 461 if (ret < 0) { 462 return -1; 463 } 464 return 0; 465 } 466 467 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt, 468 int offset, MemoryDeviceInfoList *dimms) 469 { 470 MachineState *machine = MACHINE(spapr); 471 int i, ret; 472 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE; 473 uint32_t device_lmb_start = machine->device_memory->base / lmb_size; 474 uint32_t nr_lmbs = (machine->device_memory->base + 475 memory_region_size(&machine->device_memory->mr)) / 476 lmb_size; 477 uint32_t *int_buf, *cur_index, buf_len; 478 479 /* 480 * Allocate enough buffer size to fit in ibm,dynamic-memory 481 */ 482 buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t); 483 cur_index = int_buf = g_malloc0(buf_len); 484 int_buf[0] = cpu_to_be32(nr_lmbs); 485 cur_index++; 486 for (i = 0; i < nr_lmbs; i++) { 487 uint64_t addr = i * lmb_size; 488 uint32_t *dynamic_memory = cur_index; 489 490 if (i >= device_lmb_start) { 491 SpaprDrc *drc; 492 493 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i); 494 g_assert(drc); 495 496 dynamic_memory[0] = cpu_to_be32(addr >> 32); 497 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff); 498 dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc)); 499 dynamic_memory[3] = cpu_to_be32(0); /* reserved */ 500 dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr)); 501 if (memory_region_present(get_system_memory(), addr)) { 502 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED); 503 } else { 504 dynamic_memory[5] = cpu_to_be32(0); 505 } 506 } else { 507 /* 508 * LMB information for RMA, boot time RAM and gap b/n RAM and 509 * device memory region -- all these are marked as reserved 510 * and as having no valid DRC. 511 */ 512 dynamic_memory[0] = cpu_to_be32(addr >> 32); 513 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff); 514 dynamic_memory[2] = cpu_to_be32(0); 515 dynamic_memory[3] = cpu_to_be32(0); /* reserved */ 516 dynamic_memory[4] = cpu_to_be32(-1); 517 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED | 518 SPAPR_LMB_FLAGS_DRC_INVALID); 519 } 520 521 cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE; 522 } 523 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len); 524 g_free(int_buf); 525 if (ret < 0) { 526 return -1; 527 } 528 return 0; 529 } 530 531 /* 532 * Adds ibm,dynamic-reconfiguration-memory node. 533 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation 534 * of this device tree node. 535 */ 536 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr, 537 void *fdt) 538 { 539 MachineState *machine = MACHINE(spapr); 540 int ret, offset; 541 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE; 542 uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32), 543 cpu_to_be32(lmb_size & 0xffffffff)}; 544 MemoryDeviceInfoList *dimms = NULL; 545 546 /* 547 * Don't create the node if there is no device memory 548 */ 549 if (machine->ram_size == machine->maxram_size) { 550 return 0; 551 } 552 553 offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory"); 554 555 ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size, 556 sizeof(prop_lmb_size)); 557 if (ret < 0) { 558 return ret; 559 } 560 561 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff); 562 if (ret < 0) { 563 return ret; 564 } 565 566 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0); 567 if (ret < 0) { 568 return ret; 569 } 570 571 /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */ 572 dimms = qmp_memory_device_list(); 573 if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) { 574 ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms); 575 } else { 576 ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms); 577 } 578 qapi_free_MemoryDeviceInfoList(dimms); 579 580 if (ret < 0) { 581 return ret; 582 } 583 584 ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset); 585 586 return ret; 587 } 588 589 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt) 590 { 591 MachineState *machine = MACHINE(spapr); 592 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 593 hwaddr mem_start, node_size; 594 int i, nb_nodes = machine->numa_state->num_nodes; 595 NodeInfo *nodes = machine->numa_state->nodes; 596 597 for (i = 0, mem_start = 0; i < nb_nodes; ++i) { 598 if (!nodes[i].node_mem) { 599 continue; 600 } 601 if (mem_start >= machine->ram_size) { 602 node_size = 0; 603 } else { 604 node_size = nodes[i].node_mem; 605 if (node_size > machine->ram_size - mem_start) { 606 node_size = machine->ram_size - mem_start; 607 } 608 } 609 if (!mem_start) { 610 /* spapr_machine_init() checks for rma_size <= node0_size 611 * already */ 612 spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size); 613 mem_start += spapr->rma_size; 614 node_size -= spapr->rma_size; 615 } 616 for ( ; node_size; ) { 617 hwaddr sizetmp = pow2floor(node_size); 618 619 /* mem_start != 0 here */ 620 if (ctzl(mem_start) < ctzl(sizetmp)) { 621 sizetmp = 1ULL << ctzl(mem_start); 622 } 623 624 spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp); 625 node_size -= sizetmp; 626 mem_start += sizetmp; 627 } 628 } 629 630 /* Generate ibm,dynamic-reconfiguration-memory node if required */ 631 if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) { 632 int ret; 633 634 g_assert(smc->dr_lmb_enabled); 635 ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt); 636 if (ret) { 637 return ret; 638 } 639 } 640 641 return 0; 642 } 643 644 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset, 645 SpaprMachineState *spapr) 646 { 647 MachineState *ms = MACHINE(spapr); 648 PowerPCCPU *cpu = POWERPC_CPU(cs); 649 CPUPPCState *env = &cpu->env; 650 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs); 651 int index = spapr_get_vcpu_id(cpu); 652 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40), 653 0xffffffff, 0xffffffff}; 654 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() 655 : SPAPR_TIMEBASE_FREQ; 656 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000; 657 uint32_t page_sizes_prop[64]; 658 size_t page_sizes_prop_size; 659 unsigned int smp_threads = ms->smp.threads; 660 uint32_t vcpus_per_socket = smp_threads * ms->smp.cores; 661 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)}; 662 int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu)); 663 SpaprDrc *drc; 664 int drc_index; 665 uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ]; 666 int i; 667 668 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index); 669 if (drc) { 670 drc_index = spapr_drc_index(drc); 671 _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index))); 672 } 673 674 _FDT((fdt_setprop_cell(fdt, offset, "reg", index))); 675 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu"))); 676 677 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR]))); 678 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size", 679 env->dcache_line_size))); 680 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size", 681 env->dcache_line_size))); 682 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size", 683 env->icache_line_size))); 684 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size", 685 env->icache_line_size))); 686 687 if (pcc->l1_dcache_size) { 688 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size", 689 pcc->l1_dcache_size))); 690 } else { 691 warn_report("Unknown L1 dcache size for cpu"); 692 } 693 if (pcc->l1_icache_size) { 694 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size", 695 pcc->l1_icache_size))); 696 } else { 697 warn_report("Unknown L1 icache size for cpu"); 698 } 699 700 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq))); 701 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq))); 702 _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size))); 703 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size))); 704 _FDT((fdt_setprop_string(fdt, offset, "status", "okay"))); 705 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0))); 706 707 if (ppc_has_spr(cpu, SPR_PURR)) { 708 _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1))); 709 } 710 if (ppc_has_spr(cpu, SPR_PURR)) { 711 _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1))); 712 } 713 714 if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) { 715 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes", 716 segs, sizeof(segs)))); 717 } 718 719 /* Advertise VSX (vector extensions) if available 720 * 1 == VMX / Altivec available 721 * 2 == VSX available 722 * 723 * Only CPUs for which we create core types in spapr_cpu_core.c 724 * are possible, and all of those have VMX */ 725 if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) { 726 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2))); 727 } else { 728 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1))); 729 } 730 731 /* Advertise DFP (Decimal Floating Point) if available 732 * 0 / no property == no DFP 733 * 1 == DFP available */ 734 if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) { 735 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1))); 736 } 737 738 page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop, 739 sizeof(page_sizes_prop)); 740 if (page_sizes_prop_size) { 741 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes", 742 page_sizes_prop, page_sizes_prop_size))); 743 } 744 745 spapr_dt_pa_features(spapr, cpu, fdt, offset); 746 747 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id", 748 cs->cpu_index / vcpus_per_socket))); 749 750 _FDT((fdt_setprop(fdt, offset, "ibm,pft-size", 751 pft_size_prop, sizeof(pft_size_prop)))); 752 753 if (ms->numa_state->num_nodes > 1) { 754 _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu)); 755 } 756 757 _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt)); 758 759 if (pcc->radix_page_info) { 760 for (i = 0; i < pcc->radix_page_info->count; i++) { 761 radix_AP_encodings[i] = 762 cpu_to_be32(pcc->radix_page_info->entries[i]); 763 } 764 _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings", 765 radix_AP_encodings, 766 pcc->radix_page_info->count * 767 sizeof(radix_AP_encodings[0])))); 768 } 769 770 /* 771 * We set this property to let the guest know that it can use the large 772 * decrementer and its width in bits. 773 */ 774 if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF) 775 _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits", 776 pcc->lrg_decr_bits))); 777 } 778 779 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr) 780 { 781 CPUState **rev; 782 CPUState *cs; 783 int n_cpus; 784 int cpus_offset; 785 int i; 786 787 cpus_offset = fdt_add_subnode(fdt, 0, "cpus"); 788 _FDT(cpus_offset); 789 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1))); 790 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0))); 791 792 /* 793 * We walk the CPUs in reverse order to ensure that CPU DT nodes 794 * created by fdt_add_subnode() end up in the right order in FDT 795 * for the guest kernel the enumerate the CPUs correctly. 796 * 797 * The CPU list cannot be traversed in reverse order, so we need 798 * to do extra work. 799 */ 800 n_cpus = 0; 801 rev = NULL; 802 CPU_FOREACH(cs) { 803 rev = g_renew(CPUState *, rev, n_cpus + 1); 804 rev[n_cpus++] = cs; 805 } 806 807 for (i = n_cpus - 1; i >= 0; i--) { 808 CPUState *cs = rev[i]; 809 PowerPCCPU *cpu = POWERPC_CPU(cs); 810 int index = spapr_get_vcpu_id(cpu); 811 DeviceClass *dc = DEVICE_GET_CLASS(cs); 812 g_autofree char *nodename = NULL; 813 int offset; 814 815 if (!spapr_is_thread0_in_vcore(spapr, cpu)) { 816 continue; 817 } 818 819 nodename = g_strdup_printf("%s@%x", dc->fw_name, index); 820 offset = fdt_add_subnode(fdt, cpus_offset, nodename); 821 _FDT(offset); 822 spapr_dt_cpu(cs, fdt, offset, spapr); 823 } 824 825 g_free(rev); 826 } 827 828 static int spapr_dt_rng(void *fdt) 829 { 830 int node; 831 int ret; 832 833 node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities"); 834 if (node <= 0) { 835 return -1; 836 } 837 ret = fdt_setprop_string(fdt, node, "device_type", 838 "ibm,platform-facilities"); 839 ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1); 840 ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0); 841 842 node = fdt_add_subnode(fdt, node, "ibm,random-v1"); 843 if (node <= 0) { 844 return -1; 845 } 846 ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random"); 847 848 return ret ? -1 : 0; 849 } 850 851 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt) 852 { 853 MachineState *ms = MACHINE(spapr); 854 int rtas; 855 GString *hypertas = g_string_sized_new(256); 856 GString *qemu_hypertas = g_string_sized_new(256); 857 uint64_t max_device_addr = MACHINE(spapr)->device_memory->base + 858 memory_region_size(&MACHINE(spapr)->device_memory->mr); 859 uint32_t lrdr_capacity[] = { 860 cpu_to_be32(max_device_addr >> 32), 861 cpu_to_be32(max_device_addr & 0xffffffff), 862 cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32), 863 cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff), 864 cpu_to_be32(ms->smp.max_cpus / ms->smp.threads), 865 }; 866 867 _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas")); 868 869 /* hypertas */ 870 add_str(hypertas, "hcall-pft"); 871 add_str(hypertas, "hcall-term"); 872 add_str(hypertas, "hcall-dabr"); 873 add_str(hypertas, "hcall-interrupt"); 874 add_str(hypertas, "hcall-tce"); 875 add_str(hypertas, "hcall-vio"); 876 add_str(hypertas, "hcall-splpar"); 877 add_str(hypertas, "hcall-join"); 878 add_str(hypertas, "hcall-bulk"); 879 add_str(hypertas, "hcall-set-mode"); 880 add_str(hypertas, "hcall-sprg0"); 881 add_str(hypertas, "hcall-copy"); 882 add_str(hypertas, "hcall-debug"); 883 add_str(hypertas, "hcall-vphn"); 884 if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) { 885 add_str(hypertas, "hcall-rpt-invalidate"); 886 } 887 888 add_str(qemu_hypertas, "hcall-memop1"); 889 890 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) { 891 add_str(hypertas, "hcall-multi-tce"); 892 } 893 894 if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) { 895 add_str(hypertas, "hcall-hpt-resize"); 896 } 897 898 _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions", 899 hypertas->str, hypertas->len)); 900 g_string_free(hypertas, TRUE); 901 _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions", 902 qemu_hypertas->str, qemu_hypertas->len)); 903 g_string_free(qemu_hypertas, TRUE); 904 905 spapr_numa_write_rtas_dt(spapr, fdt, rtas); 906 907 /* 908 * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log, 909 * and 16 bytes per CPU for system reset error log plus an extra 8 bytes. 910 * 911 * The system reset requirements are driven by existing Linux and PowerVM 912 * implementation which (contrary to PAPR) saves r3 in the error log 913 * structure like machine check, so Linux expects to find the saved r3 914 * value at the address in r3 upon FWNMI-enabled sreset interrupt (and 915 * does not look at the error value). 916 * 917 * System reset interrupts are not subject to interlock like machine 918 * check, so this memory area could be corrupted if the sreset is 919 * interrupted by a machine check (or vice versa) if it was shared. To 920 * prevent this, system reset uses per-CPU areas for the sreset save 921 * area. A system reset that interrupts a system reset handler could 922 * still overwrite this area, but Linux doesn't try to recover in that 923 * case anyway. 924 * 925 * The extra 8 bytes is required because Linux's FWNMI error log check 926 * is off-by-one. 927 * 928 * RTAS_MIN_SIZE is required for the RTAS blob itself. 929 */ 930 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE + 931 RTAS_ERROR_LOG_MAX + 932 ms->smp.max_cpus * sizeof(uint64_t) * 2 + 933 sizeof(uint64_t))); 934 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max", 935 RTAS_ERROR_LOG_MAX)); 936 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate", 937 RTAS_EVENT_SCAN_RATE)); 938 939 g_assert(msi_nonbroken); 940 _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0)); 941 942 /* 943 * According to PAPR, rtas ibm,os-term does not guarantee a return 944 * back to the guest cpu. 945 * 946 * While an additional ibm,extended-os-term property indicates 947 * that rtas call return will always occur. Set this property. 948 */ 949 _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0)); 950 951 _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity", 952 lrdr_capacity, sizeof(lrdr_capacity))); 953 954 spapr_dt_rtas_tokens(fdt, rtas); 955 } 956 957 /* 958 * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU 959 * and the XIVE features that the guest may request and thus the valid 960 * values for bytes 23..26 of option vector 5: 961 */ 962 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt, 963 int chosen) 964 { 965 PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu); 966 967 char val[2 * 4] = { 968 23, 0x00, /* XICS / XIVE mode */ 969 24, 0x00, /* Hash/Radix, filled in below. */ 970 25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */ 971 26, 0x40, /* Radix options: GTSE == yes. */ 972 }; 973 974 if (spapr->irq->xics && spapr->irq->xive) { 975 val[1] = SPAPR_OV5_XIVE_BOTH; 976 } else if (spapr->irq->xive) { 977 val[1] = SPAPR_OV5_XIVE_EXPLOIT; 978 } else { 979 assert(spapr->irq->xics); 980 val[1] = SPAPR_OV5_XIVE_LEGACY; 981 } 982 983 if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0, 984 first_ppc_cpu->compat_pvr)) { 985 /* 986 * If we're in a pre POWER9 compat mode then the guest should 987 * do hash and use the legacy interrupt mode 988 */ 989 val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */ 990 val[3] = 0x00; /* Hash */ 991 spapr_check_mmu_mode(false); 992 } else if (kvm_enabled()) { 993 if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) { 994 val[3] = 0x80; /* OV5_MMU_BOTH */ 995 } else if (kvmppc_has_cap_mmu_radix()) { 996 val[3] = 0x40; /* OV5_MMU_RADIX_300 */ 997 } else { 998 val[3] = 0x00; /* Hash */ 999 } 1000 } else { 1001 /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */ 1002 val[3] = 0xC0; 1003 } 1004 _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support", 1005 val, sizeof(val))); 1006 } 1007 1008 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset) 1009 { 1010 MachineState *machine = MACHINE(spapr); 1011 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine); 1012 int chosen; 1013 1014 _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen")); 1015 1016 if (reset) { 1017 const char *boot_device = spapr->boot_device; 1018 char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus); 1019 size_t cb = 0; 1020 char *bootlist = get_boot_devices_list(&cb); 1021 1022 if (machine->kernel_cmdline && machine->kernel_cmdline[0]) { 1023 _FDT(fdt_setprop_string(fdt, chosen, "bootargs", 1024 machine->kernel_cmdline)); 1025 } 1026 1027 if (spapr->initrd_size) { 1028 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start", 1029 spapr->initrd_base)); 1030 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end", 1031 spapr->initrd_base + spapr->initrd_size)); 1032 } 1033 1034 if (spapr->kernel_size) { 1035 uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr), 1036 cpu_to_be64(spapr->kernel_size) }; 1037 1038 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel", 1039 &kprop, sizeof(kprop))); 1040 if (spapr->kernel_le) { 1041 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0)); 1042 } 1043 } 1044 if (boot_menu) { 1045 _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu))); 1046 } 1047 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width)); 1048 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height)); 1049 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth)); 1050 1051 if (cb && bootlist) { 1052 int i; 1053 1054 for (i = 0; i < cb; i++) { 1055 if (bootlist[i] == '\n') { 1056 bootlist[i] = ' '; 1057 } 1058 } 1059 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist)); 1060 } 1061 1062 if (boot_device && strlen(boot_device)) { 1063 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device)); 1064 } 1065 1066 if (!spapr->has_graphics && stdout_path) { 1067 /* 1068 * "linux,stdout-path" and "stdout" properties are 1069 * deprecated by linux kernel. New platforms should only 1070 * use the "stdout-path" property. Set the new property 1071 * and continue using older property to remain compatible 1072 * with the existing firmware. 1073 */ 1074 _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path)); 1075 _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path)); 1076 } 1077 1078 /* 1079 * We can deal with BAR reallocation just fine, advertise it 1080 * to the guest 1081 */ 1082 if (smc->linux_pci_probe) { 1083 _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0)); 1084 } 1085 1086 spapr_dt_ov5_platform_support(spapr, fdt, chosen); 1087 1088 g_free(stdout_path); 1089 g_free(bootlist); 1090 } 1091 1092 _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5")); 1093 } 1094 1095 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt) 1096 { 1097 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR 1098 * KVM to work under pHyp with some guest co-operation */ 1099 int hypervisor; 1100 uint8_t hypercall[16]; 1101 1102 _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor")); 1103 /* indicate KVM hypercall interface */ 1104 _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm")); 1105 if (kvmppc_has_cap_fixup_hcalls()) { 1106 /* 1107 * Older KVM versions with older guest kernels were broken 1108 * with the magic page, don't allow the guest to map it. 1109 */ 1110 if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall, 1111 sizeof(hypercall))) { 1112 _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions", 1113 hypercall, sizeof(hypercall))); 1114 } 1115 } 1116 } 1117 1118 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space) 1119 { 1120 MachineState *machine = MACHINE(spapr); 1121 MachineClass *mc = MACHINE_GET_CLASS(machine); 1122 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine); 1123 uint32_t root_drc_type_mask = 0; 1124 int ret; 1125 void *fdt; 1126 SpaprPhbState *phb; 1127 char *buf; 1128 1129 fdt = g_malloc0(space); 1130 _FDT((fdt_create_empty_tree(fdt, space))); 1131 1132 /* Root node */ 1133 _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp")); 1134 _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)")); 1135 _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries")); 1136 1137 /* Guest UUID & Name*/ 1138 buf = qemu_uuid_unparse_strdup(&qemu_uuid); 1139 _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf)); 1140 if (qemu_uuid_set) { 1141 _FDT(fdt_setprop_string(fdt, 0, "system-id", buf)); 1142 } 1143 g_free(buf); 1144 1145 if (qemu_get_vm_name()) { 1146 _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name", 1147 qemu_get_vm_name())); 1148 } 1149 1150 /* Host Model & Serial Number */ 1151 if (spapr->host_model) { 1152 _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model)); 1153 } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) { 1154 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf)); 1155 g_free(buf); 1156 } 1157 1158 if (spapr->host_serial) { 1159 _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial)); 1160 } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) { 1161 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf)); 1162 g_free(buf); 1163 } 1164 1165 _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2)); 1166 _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2)); 1167 1168 /* /interrupt controller */ 1169 spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC); 1170 1171 ret = spapr_dt_memory(spapr, fdt); 1172 if (ret < 0) { 1173 error_report("couldn't setup memory nodes in fdt"); 1174 exit(1); 1175 } 1176 1177 /* /vdevice */ 1178 spapr_dt_vdevice(spapr->vio_bus, fdt); 1179 1180 if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) { 1181 ret = spapr_dt_rng(fdt); 1182 if (ret < 0) { 1183 error_report("could not set up rng device in the fdt"); 1184 exit(1); 1185 } 1186 } 1187 1188 QLIST_FOREACH(phb, &spapr->phbs, list) { 1189 ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL); 1190 if (ret < 0) { 1191 error_report("couldn't setup PCI devices in fdt"); 1192 exit(1); 1193 } 1194 } 1195 1196 spapr_dt_cpus(fdt, spapr); 1197 1198 /* ibm,drc-indexes and friends */ 1199 if (smc->dr_lmb_enabled) { 1200 root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB; 1201 } 1202 if (smc->dr_phb_enabled) { 1203 root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB; 1204 } 1205 if (mc->nvdimm_supported) { 1206 root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM; 1207 } 1208 if (root_drc_type_mask) { 1209 _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask)); 1210 } 1211 1212 if (mc->has_hotpluggable_cpus) { 1213 int offset = fdt_path_offset(fdt, "/cpus"); 1214 ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU); 1215 if (ret < 0) { 1216 error_report("Couldn't set up CPU DR device tree properties"); 1217 exit(1); 1218 } 1219 } 1220 1221 /* /event-sources */ 1222 spapr_dt_events(spapr, fdt); 1223 1224 /* /rtas */ 1225 spapr_dt_rtas(spapr, fdt); 1226 1227 /* /chosen */ 1228 spapr_dt_chosen(spapr, fdt, reset); 1229 1230 /* /hypervisor */ 1231 if (kvm_enabled()) { 1232 spapr_dt_hypervisor(spapr, fdt); 1233 } 1234 1235 /* Build memory reserve map */ 1236 if (reset) { 1237 if (spapr->kernel_size) { 1238 _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr, 1239 spapr->kernel_size))); 1240 } 1241 if (spapr->initrd_size) { 1242 _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, 1243 spapr->initrd_size))); 1244 } 1245 } 1246 1247 /* NVDIMM devices */ 1248 if (mc->nvdimm_supported) { 1249 spapr_dt_persistent_memory(spapr, fdt); 1250 } 1251 1252 return fdt; 1253 } 1254 1255 static uint64_t translate_kernel_address(void *opaque, uint64_t addr) 1256 { 1257 SpaprMachineState *spapr = opaque; 1258 1259 return (addr & 0x0fffffff) + spapr->kernel_addr; 1260 } 1261 1262 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp, 1263 PowerPCCPU *cpu) 1264 { 1265 CPUPPCState *env = &cpu->env; 1266 1267 /* The TCG path should also be holding the BQL at this point */ 1268 g_assert(qemu_mutex_iothread_locked()); 1269 1270 if (msr_pr) { 1271 hcall_dprintf("Hypercall made with MSR[PR]=1\n"); 1272 env->gpr[3] = H_PRIVILEGE; 1273 } else { 1274 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]); 1275 } 1276 } 1277 1278 struct LPCRSyncState { 1279 target_ulong value; 1280 target_ulong mask; 1281 }; 1282 1283 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg) 1284 { 1285 struct LPCRSyncState *s = arg.host_ptr; 1286 PowerPCCPU *cpu = POWERPC_CPU(cs); 1287 CPUPPCState *env = &cpu->env; 1288 target_ulong lpcr; 1289 1290 cpu_synchronize_state(cs); 1291 lpcr = env->spr[SPR_LPCR]; 1292 lpcr &= ~s->mask; 1293 lpcr |= s->value; 1294 ppc_store_lpcr(cpu, lpcr); 1295 } 1296 1297 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask) 1298 { 1299 CPUState *cs; 1300 struct LPCRSyncState s = { 1301 .value = value, 1302 .mask = mask 1303 }; 1304 CPU_FOREACH(cs) { 1305 run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s)); 1306 } 1307 } 1308 1309 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry) 1310 { 1311 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp); 1312 1313 /* Copy PATE1:GR into PATE0:HR */ 1314 entry->dw0 = spapr->patb_entry & PATE0_HR; 1315 entry->dw1 = spapr->patb_entry; 1316 } 1317 1318 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2)) 1319 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID) 1320 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY) 1321 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY)) 1322 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY)) 1323 1324 /* 1325 * Get the fd to access the kernel htab, re-opening it if necessary 1326 */ 1327 static int get_htab_fd(SpaprMachineState *spapr) 1328 { 1329 Error *local_err = NULL; 1330 1331 if (spapr->htab_fd >= 0) { 1332 return spapr->htab_fd; 1333 } 1334 1335 spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err); 1336 if (spapr->htab_fd < 0) { 1337 error_report_err(local_err); 1338 } 1339 1340 return spapr->htab_fd; 1341 } 1342 1343 void close_htab_fd(SpaprMachineState *spapr) 1344 { 1345 if (spapr->htab_fd >= 0) { 1346 close(spapr->htab_fd); 1347 } 1348 spapr->htab_fd = -1; 1349 } 1350 1351 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp) 1352 { 1353 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp); 1354 1355 return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1; 1356 } 1357 1358 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp) 1359 { 1360 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp); 1361 1362 assert(kvm_enabled()); 1363 1364 if (!spapr->htab) { 1365 return 0; 1366 } 1367 1368 return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18); 1369 } 1370 1371 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp, 1372 hwaddr ptex, int n) 1373 { 1374 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp); 1375 hwaddr pte_offset = ptex * HASH_PTE_SIZE_64; 1376 1377 if (!spapr->htab) { 1378 /* 1379 * HTAB is controlled by KVM. Fetch into temporary buffer 1380 */ 1381 ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64); 1382 kvmppc_read_hptes(hptes, ptex, n); 1383 return hptes; 1384 } 1385 1386 /* 1387 * HTAB is controlled by QEMU. Just point to the internally 1388 * accessible PTEG. 1389 */ 1390 return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset); 1391 } 1392 1393 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp, 1394 const ppc_hash_pte64_t *hptes, 1395 hwaddr ptex, int n) 1396 { 1397 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp); 1398 1399 if (!spapr->htab) { 1400 g_free((void *)hptes); 1401 } 1402 1403 /* Nothing to do for qemu managed HPT */ 1404 } 1405 1406 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex, 1407 uint64_t pte0, uint64_t pte1) 1408 { 1409 SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp); 1410 hwaddr offset = ptex * HASH_PTE_SIZE_64; 1411 1412 if (!spapr->htab) { 1413 kvmppc_write_hpte(ptex, pte0, pte1); 1414 } else { 1415 if (pte0 & HPTE64_V_VALID) { 1416 stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1); 1417 /* 1418 * When setting valid, we write PTE1 first. This ensures 1419 * proper synchronization with the reading code in 1420 * ppc_hash64_pteg_search() 1421 */ 1422 smp_wmb(); 1423 stq_p(spapr->htab + offset, pte0); 1424 } else { 1425 stq_p(spapr->htab + offset, pte0); 1426 /* 1427 * When clearing it we set PTE0 first. This ensures proper 1428 * synchronization with the reading code in 1429 * ppc_hash64_pteg_search() 1430 */ 1431 smp_wmb(); 1432 stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1); 1433 } 1434 } 1435 } 1436 1437 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex, 1438 uint64_t pte1) 1439 { 1440 hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15; 1441 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp); 1442 1443 if (!spapr->htab) { 1444 /* There should always be a hash table when this is called */ 1445 error_report("spapr_hpte_set_c called with no hash table !"); 1446 return; 1447 } 1448 1449 /* The HW performs a non-atomic byte update */ 1450 stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80); 1451 } 1452 1453 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex, 1454 uint64_t pte1) 1455 { 1456 hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14; 1457 SpaprMachineState *spapr = SPAPR_MACHINE(vhyp); 1458 1459 if (!spapr->htab) { 1460 /* There should always be a hash table when this is called */ 1461 error_report("spapr_hpte_set_r called with no hash table !"); 1462 return; 1463 } 1464 1465 /* The HW performs a non-atomic byte update */ 1466 stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01); 1467 } 1468 1469 int spapr_hpt_shift_for_ramsize(uint64_t ramsize) 1470 { 1471 int shift; 1472 1473 /* We aim for a hash table of size 1/128 the size of RAM (rounded 1474 * up). The PAPR recommendation is actually 1/64 of RAM size, but 1475 * that's much more than is needed for Linux guests */ 1476 shift = ctz64(pow2ceil(ramsize)) - 7; 1477 shift = MAX(shift, 18); /* Minimum architected size */ 1478 shift = MIN(shift, 46); /* Maximum architected size */ 1479 return shift; 1480 } 1481 1482 void spapr_free_hpt(SpaprMachineState *spapr) 1483 { 1484 g_free(spapr->htab); 1485 spapr->htab = NULL; 1486 spapr->htab_shift = 0; 1487 close_htab_fd(spapr); 1488 } 1489 1490 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp) 1491 { 1492 ERRP_GUARD(); 1493 long rc; 1494 1495 /* Clean up any HPT info from a previous boot */ 1496 spapr_free_hpt(spapr); 1497 1498 rc = kvmppc_reset_htab(shift); 1499 1500 if (rc == -EOPNOTSUPP) { 1501 error_setg(errp, "HPT not supported in nested guests"); 1502 return -EOPNOTSUPP; 1503 } 1504 1505 if (rc < 0) { 1506 /* kernel-side HPT needed, but couldn't allocate one */ 1507 error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d", 1508 shift); 1509 error_append_hint(errp, "Try smaller maxmem?\n"); 1510 return -errno; 1511 } else if (rc > 0) { 1512 /* kernel-side HPT allocated */ 1513 if (rc != shift) { 1514 error_setg(errp, 1515 "Requested order %d HPT, but kernel allocated order %ld", 1516 shift, rc); 1517 error_append_hint(errp, "Try smaller maxmem?\n"); 1518 return -ENOSPC; 1519 } 1520 1521 spapr->htab_shift = shift; 1522 spapr->htab = NULL; 1523 } else { 1524 /* kernel-side HPT not needed, allocate in userspace instead */ 1525 size_t size = 1ULL << shift; 1526 int i; 1527 1528 spapr->htab = qemu_memalign(size, size); 1529 memset(spapr->htab, 0, size); 1530 spapr->htab_shift = shift; 1531 1532 for (i = 0; i < size / HASH_PTE_SIZE_64; i++) { 1533 DIRTY_HPTE(HPTE(spapr->htab, i)); 1534 } 1535 } 1536 /* We're setting up a hash table, so that means we're not radix */ 1537 spapr->patb_entry = 0; 1538 spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT); 1539 return 0; 1540 } 1541 1542 void spapr_setup_hpt(SpaprMachineState *spapr) 1543 { 1544 int hpt_shift; 1545 1546 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) { 1547 hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size); 1548 } else { 1549 uint64_t current_ram_size; 1550 1551 current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size(); 1552 hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size); 1553 } 1554 spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal); 1555 1556 if (kvm_enabled()) { 1557 hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift); 1558 1559 /* Check our RMA fits in the possible VRMA */ 1560 if (vrma_limit < spapr->rma_size) { 1561 error_report("Unable to create %" HWADDR_PRIu 1562 "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB", 1563 spapr->rma_size / MiB, vrma_limit / MiB); 1564 exit(EXIT_FAILURE); 1565 } 1566 } 1567 } 1568 1569 void spapr_check_mmu_mode(bool guest_radix) 1570 { 1571 if (guest_radix) { 1572 if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) { 1573 error_report("Guest requested unavailable MMU mode (radix)."); 1574 exit(EXIT_FAILURE); 1575 } 1576 } else { 1577 if (kvm_enabled() && kvmppc_has_cap_mmu_radix() 1578 && !kvmppc_has_cap_mmu_hash_v3()) { 1579 error_report("Guest requested unavailable MMU mode (hash)."); 1580 exit(EXIT_FAILURE); 1581 } 1582 } 1583 } 1584 1585 static void spapr_machine_reset(MachineState *machine) 1586 { 1587 SpaprMachineState *spapr = SPAPR_MACHINE(machine); 1588 PowerPCCPU *first_ppc_cpu; 1589 hwaddr fdt_addr; 1590 void *fdt; 1591 int rc; 1592 1593 pef_kvm_reset(machine->cgs, &error_fatal); 1594 spapr_caps_apply(spapr); 1595 1596 first_ppc_cpu = POWERPC_CPU(first_cpu); 1597 if (kvm_enabled() && kvmppc_has_cap_mmu_radix() && 1598 ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0, 1599 spapr->max_compat_pvr)) { 1600 /* 1601 * If using KVM with radix mode available, VCPUs can be started 1602 * without a HPT because KVM will start them in radix mode. 1603 * Set the GR bit in PATE so that we know there is no HPT. 1604 */ 1605 spapr->patb_entry = PATE1_GR; 1606 spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT); 1607 } else { 1608 spapr_setup_hpt(spapr); 1609 } 1610 1611 qemu_devices_reset(); 1612 1613 spapr_ovec_cleanup(spapr->ov5_cas); 1614 spapr->ov5_cas = spapr_ovec_new(); 1615 1616 ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal); 1617 1618 /* 1619 * This is fixing some of the default configuration of the XIVE 1620 * devices. To be called after the reset of the machine devices. 1621 */ 1622 spapr_irq_reset(spapr, &error_fatal); 1623 1624 /* 1625 * There is no CAS under qtest. Simulate one to please the code that 1626 * depends on spapr->ov5_cas. This is especially needed to test device 1627 * unplug, so we do that before resetting the DRCs. 1628 */ 1629 if (qtest_enabled()) { 1630 spapr_ovec_cleanup(spapr->ov5_cas); 1631 spapr->ov5_cas = spapr_ovec_clone(spapr->ov5); 1632 } 1633 1634 /* DRC reset may cause a device to be unplugged. This will cause troubles 1635 * if this device is used by another device (eg, a running vhost backend 1636 * will crash QEMU if the DIMM holding the vring goes away). To avoid such 1637 * situations, we reset DRCs after all devices have been reset. 1638 */ 1639 spapr_drc_reset_all(spapr); 1640 1641 spapr_clear_pending_events(spapr); 1642 1643 /* 1644 * We place the device tree just below either the top of the RMA, 1645 * or just below 2GB, whichever is lower, so that it can be 1646 * processed with 32-bit real mode code if necessary 1647 */ 1648 fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE; 1649 1650 fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE); 1651 if (spapr->vof) { 1652 spapr_vof_reset(spapr, fdt, &error_fatal); 1653 /* 1654 * Do not pack the FDT as the client may change properties. 1655 * VOF client does not expect the FDT so we do not load it to the VM. 1656 */ 1657 } else { 1658 rc = fdt_pack(fdt); 1659 /* Should only fail if we've built a corrupted tree */ 1660 assert(rc == 0); 1661 1662 spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 1663 0, fdt_addr, 0); 1664 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt)); 1665 } 1666 qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt)); 1667 1668 g_free(spapr->fdt_blob); 1669 spapr->fdt_size = fdt_totalsize(fdt); 1670 spapr->fdt_initial_size = spapr->fdt_size; 1671 spapr->fdt_blob = fdt; 1672 1673 /* Set up the entry state */ 1674 first_ppc_cpu->env.gpr[5] = 0; 1675 1676 spapr->fwnmi_system_reset_addr = -1; 1677 spapr->fwnmi_machine_check_addr = -1; 1678 spapr->fwnmi_machine_check_interlock = -1; 1679 1680 /* Signal all vCPUs waiting on this condition */ 1681 qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond); 1682 1683 migrate_del_blocker(spapr->fwnmi_migration_blocker); 1684 } 1685 1686 static void spapr_create_nvram(SpaprMachineState *spapr) 1687 { 1688 DeviceState *dev = qdev_new("spapr-nvram"); 1689 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0); 1690 1691 if (dinfo) { 1692 qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo), 1693 &error_fatal); 1694 } 1695 1696 qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal); 1697 1698 spapr->nvram = (struct SpaprNvram *)dev; 1699 } 1700 1701 static void spapr_rtc_create(SpaprMachineState *spapr) 1702 { 1703 object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc, 1704 sizeof(spapr->rtc), TYPE_SPAPR_RTC, 1705 &error_fatal, NULL); 1706 qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal); 1707 object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc), 1708 "date"); 1709 } 1710 1711 /* Returns whether we want to use VGA or not */ 1712 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp) 1713 { 1714 switch (vga_interface_type) { 1715 case VGA_NONE: 1716 return false; 1717 case VGA_DEVICE: 1718 return true; 1719 case VGA_STD: 1720 case VGA_VIRTIO: 1721 case VGA_CIRRUS: 1722 return pci_vga_init(pci_bus) != NULL; 1723 default: 1724 error_setg(errp, 1725 "Unsupported VGA mode, only -vga std or -vga virtio is supported"); 1726 return false; 1727 } 1728 } 1729 1730 static int spapr_pre_load(void *opaque) 1731 { 1732 int rc; 1733 1734 rc = spapr_caps_pre_load(opaque); 1735 if (rc) { 1736 return rc; 1737 } 1738 1739 return 0; 1740 } 1741 1742 static int spapr_post_load(void *opaque, int version_id) 1743 { 1744 SpaprMachineState *spapr = (SpaprMachineState *)opaque; 1745 int err = 0; 1746 1747 err = spapr_caps_post_migration(spapr); 1748 if (err) { 1749 return err; 1750 } 1751 1752 /* 1753 * In earlier versions, there was no separate qdev for the PAPR 1754 * RTC, so the RTC offset was stored directly in sPAPREnvironment. 1755 * So when migrating from those versions, poke the incoming offset 1756 * value into the RTC device 1757 */ 1758 if (version_id < 3) { 1759 err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset); 1760 if (err) { 1761 return err; 1762 } 1763 } 1764 1765 if (kvm_enabled() && spapr->patb_entry) { 1766 PowerPCCPU *cpu = POWERPC_CPU(first_cpu); 1767 bool radix = !!(spapr->patb_entry & PATE1_GR); 1768 bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE); 1769 1770 /* 1771 * Update LPCR:HR and UPRT as they may not be set properly in 1772 * the stream 1773 */ 1774 spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0, 1775 LPCR_HR | LPCR_UPRT); 1776 1777 err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry); 1778 if (err) { 1779 error_report("Process table config unsupported by the host"); 1780 return -EINVAL; 1781 } 1782 } 1783 1784 err = spapr_irq_post_load(spapr, version_id); 1785 if (err) { 1786 return err; 1787 } 1788 1789 return err; 1790 } 1791 1792 static int spapr_pre_save(void *opaque) 1793 { 1794 int rc; 1795 1796 rc = spapr_caps_pre_save(opaque); 1797 if (rc) { 1798 return rc; 1799 } 1800 1801 return 0; 1802 } 1803 1804 static bool version_before_3(void *opaque, int version_id) 1805 { 1806 return version_id < 3; 1807 } 1808 1809 static bool spapr_pending_events_needed(void *opaque) 1810 { 1811 SpaprMachineState *spapr = (SpaprMachineState *)opaque; 1812 return !QTAILQ_EMPTY(&spapr->pending_events); 1813 } 1814 1815 static const VMStateDescription vmstate_spapr_event_entry = { 1816 .name = "spapr_event_log_entry", 1817 .version_id = 1, 1818 .minimum_version_id = 1, 1819 .fields = (VMStateField[]) { 1820 VMSTATE_UINT32(summary, SpaprEventLogEntry), 1821 VMSTATE_UINT32(extended_length, SpaprEventLogEntry), 1822 VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0, 1823 NULL, extended_length), 1824 VMSTATE_END_OF_LIST() 1825 }, 1826 }; 1827 1828 static const VMStateDescription vmstate_spapr_pending_events = { 1829 .name = "spapr_pending_events", 1830 .version_id = 1, 1831 .minimum_version_id = 1, 1832 .needed = spapr_pending_events_needed, 1833 .fields = (VMStateField[]) { 1834 VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1, 1835 vmstate_spapr_event_entry, SpaprEventLogEntry, next), 1836 VMSTATE_END_OF_LIST() 1837 }, 1838 }; 1839 1840 static bool spapr_ov5_cas_needed(void *opaque) 1841 { 1842 SpaprMachineState *spapr = opaque; 1843 SpaprOptionVector *ov5_mask = spapr_ovec_new(); 1844 bool cas_needed; 1845 1846 /* Prior to the introduction of SpaprOptionVector, we had two option 1847 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY. 1848 * Both of these options encode machine topology into the device-tree 1849 * in such a way that the now-booted OS should still be able to interact 1850 * appropriately with QEMU regardless of what options were actually 1851 * negotiatied on the source side. 1852 * 1853 * As such, we can avoid migrating the CAS-negotiated options if these 1854 * are the only options available on the current machine/platform. 1855 * Since these are the only options available for pseries-2.7 and 1856 * earlier, this allows us to maintain old->new/new->old migration 1857 * compatibility. 1858 * 1859 * For QEMU 2.8+, there are additional CAS-negotiatable options available 1860 * via default pseries-2.8 machines and explicit command-line parameters. 1861 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware 1862 * of the actual CAS-negotiated values to continue working properly. For 1863 * example, availability of memory unplug depends on knowing whether 1864 * OV5_HP_EVT was negotiated via CAS. 1865 * 1866 * Thus, for any cases where the set of available CAS-negotiatable 1867 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we 1868 * include the CAS-negotiated options in the migration stream, unless 1869 * if they affect boot time behaviour only. 1870 */ 1871 spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY); 1872 spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY); 1873 spapr_ovec_set(ov5_mask, OV5_DRMEM_V2); 1874 1875 /* We need extra information if we have any bits outside the mask 1876 * defined above */ 1877 cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask); 1878 1879 spapr_ovec_cleanup(ov5_mask); 1880 1881 return cas_needed; 1882 } 1883 1884 static const VMStateDescription vmstate_spapr_ov5_cas = { 1885 .name = "spapr_option_vector_ov5_cas", 1886 .version_id = 1, 1887 .minimum_version_id = 1, 1888 .needed = spapr_ov5_cas_needed, 1889 .fields = (VMStateField[]) { 1890 VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1, 1891 vmstate_spapr_ovec, SpaprOptionVector), 1892 VMSTATE_END_OF_LIST() 1893 }, 1894 }; 1895 1896 static bool spapr_patb_entry_needed(void *opaque) 1897 { 1898 SpaprMachineState *spapr = opaque; 1899 1900 return !!spapr->patb_entry; 1901 } 1902 1903 static const VMStateDescription vmstate_spapr_patb_entry = { 1904 .name = "spapr_patb_entry", 1905 .version_id = 1, 1906 .minimum_version_id = 1, 1907 .needed = spapr_patb_entry_needed, 1908 .fields = (VMStateField[]) { 1909 VMSTATE_UINT64(patb_entry, SpaprMachineState), 1910 VMSTATE_END_OF_LIST() 1911 }, 1912 }; 1913 1914 static bool spapr_irq_map_needed(void *opaque) 1915 { 1916 SpaprMachineState *spapr = opaque; 1917 1918 return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr); 1919 } 1920 1921 static const VMStateDescription vmstate_spapr_irq_map = { 1922 .name = "spapr_irq_map", 1923 .version_id = 1, 1924 .minimum_version_id = 1, 1925 .needed = spapr_irq_map_needed, 1926 .fields = (VMStateField[]) { 1927 VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr), 1928 VMSTATE_END_OF_LIST() 1929 }, 1930 }; 1931 1932 static bool spapr_dtb_needed(void *opaque) 1933 { 1934 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque); 1935 1936 return smc->update_dt_enabled; 1937 } 1938 1939 static int spapr_dtb_pre_load(void *opaque) 1940 { 1941 SpaprMachineState *spapr = (SpaprMachineState *)opaque; 1942 1943 g_free(spapr->fdt_blob); 1944 spapr->fdt_blob = NULL; 1945 spapr->fdt_size = 0; 1946 1947 return 0; 1948 } 1949 1950 static const VMStateDescription vmstate_spapr_dtb = { 1951 .name = "spapr_dtb", 1952 .version_id = 1, 1953 .minimum_version_id = 1, 1954 .needed = spapr_dtb_needed, 1955 .pre_load = spapr_dtb_pre_load, 1956 .fields = (VMStateField[]) { 1957 VMSTATE_UINT32(fdt_initial_size, SpaprMachineState), 1958 VMSTATE_UINT32(fdt_size, SpaprMachineState), 1959 VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL, 1960 fdt_size), 1961 VMSTATE_END_OF_LIST() 1962 }, 1963 }; 1964 1965 static bool spapr_fwnmi_needed(void *opaque) 1966 { 1967 SpaprMachineState *spapr = (SpaprMachineState *)opaque; 1968 1969 return spapr->fwnmi_machine_check_addr != -1; 1970 } 1971 1972 static int spapr_fwnmi_pre_save(void *opaque) 1973 { 1974 SpaprMachineState *spapr = (SpaprMachineState *)opaque; 1975 1976 /* 1977 * Check if machine check handling is in progress and print a 1978 * warning message. 1979 */ 1980 if (spapr->fwnmi_machine_check_interlock != -1) { 1981 warn_report("A machine check is being handled during migration. The" 1982 "handler may run and log hardware error on the destination"); 1983 } 1984 1985 return 0; 1986 } 1987 1988 static const VMStateDescription vmstate_spapr_fwnmi = { 1989 .name = "spapr_fwnmi", 1990 .version_id = 1, 1991 .minimum_version_id = 1, 1992 .needed = spapr_fwnmi_needed, 1993 .pre_save = spapr_fwnmi_pre_save, 1994 .fields = (VMStateField[]) { 1995 VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState), 1996 VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState), 1997 VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState), 1998 VMSTATE_END_OF_LIST() 1999 }, 2000 }; 2001 2002 static const VMStateDescription vmstate_spapr = { 2003 .name = "spapr", 2004 .version_id = 3, 2005 .minimum_version_id = 1, 2006 .pre_load = spapr_pre_load, 2007 .post_load = spapr_post_load, 2008 .pre_save = spapr_pre_save, 2009 .fields = (VMStateField[]) { 2010 /* used to be @next_irq */ 2011 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4), 2012 2013 /* RTC offset */ 2014 VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3), 2015 2016 VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2), 2017 VMSTATE_END_OF_LIST() 2018 }, 2019 .subsections = (const VMStateDescription*[]) { 2020 &vmstate_spapr_ov5_cas, 2021 &vmstate_spapr_patb_entry, 2022 &vmstate_spapr_pending_events, 2023 &vmstate_spapr_cap_htm, 2024 &vmstate_spapr_cap_vsx, 2025 &vmstate_spapr_cap_dfp, 2026 &vmstate_spapr_cap_cfpc, 2027 &vmstate_spapr_cap_sbbc, 2028 &vmstate_spapr_cap_ibs, 2029 &vmstate_spapr_cap_hpt_maxpagesize, 2030 &vmstate_spapr_irq_map, 2031 &vmstate_spapr_cap_nested_kvm_hv, 2032 &vmstate_spapr_dtb, 2033 &vmstate_spapr_cap_large_decr, 2034 &vmstate_spapr_cap_ccf_assist, 2035 &vmstate_spapr_cap_fwnmi, 2036 &vmstate_spapr_fwnmi, 2037 &vmstate_spapr_cap_rpt_invalidate, 2038 NULL 2039 } 2040 }; 2041 2042 static int htab_save_setup(QEMUFile *f, void *opaque) 2043 { 2044 SpaprMachineState *spapr = opaque; 2045 2046 /* "Iteration" header */ 2047 if (!spapr->htab_shift) { 2048 qemu_put_be32(f, -1); 2049 } else { 2050 qemu_put_be32(f, spapr->htab_shift); 2051 } 2052 2053 if (spapr->htab) { 2054 spapr->htab_save_index = 0; 2055 spapr->htab_first_pass = true; 2056 } else { 2057 if (spapr->htab_shift) { 2058 assert(kvm_enabled()); 2059 } 2060 } 2061 2062 2063 return 0; 2064 } 2065 2066 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr, 2067 int chunkstart, int n_valid, int n_invalid) 2068 { 2069 qemu_put_be32(f, chunkstart); 2070 qemu_put_be16(f, n_valid); 2071 qemu_put_be16(f, n_invalid); 2072 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart), 2073 HASH_PTE_SIZE_64 * n_valid); 2074 } 2075 2076 static void htab_save_end_marker(QEMUFile *f) 2077 { 2078 qemu_put_be32(f, 0); 2079 qemu_put_be16(f, 0); 2080 qemu_put_be16(f, 0); 2081 } 2082 2083 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr, 2084 int64_t max_ns) 2085 { 2086 bool has_timeout = max_ns != -1; 2087 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64; 2088 int index = spapr->htab_save_index; 2089 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 2090 2091 assert(spapr->htab_first_pass); 2092 2093 do { 2094 int chunkstart; 2095 2096 /* Consume invalid HPTEs */ 2097 while ((index < htabslots) 2098 && !HPTE_VALID(HPTE(spapr->htab, index))) { 2099 CLEAN_HPTE(HPTE(spapr->htab, index)); 2100 index++; 2101 } 2102 2103 /* Consume valid HPTEs */ 2104 chunkstart = index; 2105 while ((index < htabslots) && (index - chunkstart < USHRT_MAX) 2106 && HPTE_VALID(HPTE(spapr->htab, index))) { 2107 CLEAN_HPTE(HPTE(spapr->htab, index)); 2108 index++; 2109 } 2110 2111 if (index > chunkstart) { 2112 int n_valid = index - chunkstart; 2113 2114 htab_save_chunk(f, spapr, chunkstart, n_valid, 0); 2115 2116 if (has_timeout && 2117 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) { 2118 break; 2119 } 2120 } 2121 } while ((index < htabslots) && !qemu_file_rate_limit(f)); 2122 2123 if (index >= htabslots) { 2124 assert(index == htabslots); 2125 index = 0; 2126 spapr->htab_first_pass = false; 2127 } 2128 spapr->htab_save_index = index; 2129 } 2130 2131 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr, 2132 int64_t max_ns) 2133 { 2134 bool final = max_ns < 0; 2135 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64; 2136 int examined = 0, sent = 0; 2137 int index = spapr->htab_save_index; 2138 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 2139 2140 assert(!spapr->htab_first_pass); 2141 2142 do { 2143 int chunkstart, invalidstart; 2144 2145 /* Consume non-dirty HPTEs */ 2146 while ((index < htabslots) 2147 && !HPTE_DIRTY(HPTE(spapr->htab, index))) { 2148 index++; 2149 examined++; 2150 } 2151 2152 chunkstart = index; 2153 /* Consume valid dirty HPTEs */ 2154 while ((index < htabslots) && (index - chunkstart < USHRT_MAX) 2155 && HPTE_DIRTY(HPTE(spapr->htab, index)) 2156 && HPTE_VALID(HPTE(spapr->htab, index))) { 2157 CLEAN_HPTE(HPTE(spapr->htab, index)); 2158 index++; 2159 examined++; 2160 } 2161 2162 invalidstart = index; 2163 /* Consume invalid dirty HPTEs */ 2164 while ((index < htabslots) && (index - invalidstart < USHRT_MAX) 2165 && HPTE_DIRTY(HPTE(spapr->htab, index)) 2166 && !HPTE_VALID(HPTE(spapr->htab, index))) { 2167 CLEAN_HPTE(HPTE(spapr->htab, index)); 2168 index++; 2169 examined++; 2170 } 2171 2172 if (index > chunkstart) { 2173 int n_valid = invalidstart - chunkstart; 2174 int n_invalid = index - invalidstart; 2175 2176 htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid); 2177 sent += index - chunkstart; 2178 2179 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) { 2180 break; 2181 } 2182 } 2183 2184 if (examined >= htabslots) { 2185 break; 2186 } 2187 2188 if (index >= htabslots) { 2189 assert(index == htabslots); 2190 index = 0; 2191 } 2192 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final)); 2193 2194 if (index >= htabslots) { 2195 assert(index == htabslots); 2196 index = 0; 2197 } 2198 2199 spapr->htab_save_index = index; 2200 2201 return (examined >= htabslots) && (sent == 0) ? 1 : 0; 2202 } 2203 2204 #define MAX_ITERATION_NS 5000000 /* 5 ms */ 2205 #define MAX_KVM_BUF_SIZE 2048 2206 2207 static int htab_save_iterate(QEMUFile *f, void *opaque) 2208 { 2209 SpaprMachineState *spapr = opaque; 2210 int fd; 2211 int rc = 0; 2212 2213 /* Iteration header */ 2214 if (!spapr->htab_shift) { 2215 qemu_put_be32(f, -1); 2216 return 1; 2217 } else { 2218 qemu_put_be32(f, 0); 2219 } 2220 2221 if (!spapr->htab) { 2222 assert(kvm_enabled()); 2223 2224 fd = get_htab_fd(spapr); 2225 if (fd < 0) { 2226 return fd; 2227 } 2228 2229 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS); 2230 if (rc < 0) { 2231 return rc; 2232 } 2233 } else if (spapr->htab_first_pass) { 2234 htab_save_first_pass(f, spapr, MAX_ITERATION_NS); 2235 } else { 2236 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS); 2237 } 2238 2239 htab_save_end_marker(f); 2240 2241 return rc; 2242 } 2243 2244 static int htab_save_complete(QEMUFile *f, void *opaque) 2245 { 2246 SpaprMachineState *spapr = opaque; 2247 int fd; 2248 2249 /* Iteration header */ 2250 if (!spapr->htab_shift) { 2251 qemu_put_be32(f, -1); 2252 return 0; 2253 } else { 2254 qemu_put_be32(f, 0); 2255 } 2256 2257 if (!spapr->htab) { 2258 int rc; 2259 2260 assert(kvm_enabled()); 2261 2262 fd = get_htab_fd(spapr); 2263 if (fd < 0) { 2264 return fd; 2265 } 2266 2267 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1); 2268 if (rc < 0) { 2269 return rc; 2270 } 2271 } else { 2272 if (spapr->htab_first_pass) { 2273 htab_save_first_pass(f, spapr, -1); 2274 } 2275 htab_save_later_pass(f, spapr, -1); 2276 } 2277 2278 /* End marker */ 2279 htab_save_end_marker(f); 2280 2281 return 0; 2282 } 2283 2284 static int htab_load(QEMUFile *f, void *opaque, int version_id) 2285 { 2286 SpaprMachineState *spapr = opaque; 2287 uint32_t section_hdr; 2288 int fd = -1; 2289 Error *local_err = NULL; 2290 2291 if (version_id < 1 || version_id > 1) { 2292 error_report("htab_load() bad version"); 2293 return -EINVAL; 2294 } 2295 2296 section_hdr = qemu_get_be32(f); 2297 2298 if (section_hdr == -1) { 2299 spapr_free_hpt(spapr); 2300 return 0; 2301 } 2302 2303 if (section_hdr) { 2304 int ret; 2305 2306 /* First section gives the htab size */ 2307 ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err); 2308 if (ret < 0) { 2309 error_report_err(local_err); 2310 return ret; 2311 } 2312 return 0; 2313 } 2314 2315 if (!spapr->htab) { 2316 assert(kvm_enabled()); 2317 2318 fd = kvmppc_get_htab_fd(true, 0, &local_err); 2319 if (fd < 0) { 2320 error_report_err(local_err); 2321 return fd; 2322 } 2323 } 2324 2325 while (true) { 2326 uint32_t index; 2327 uint16_t n_valid, n_invalid; 2328 2329 index = qemu_get_be32(f); 2330 n_valid = qemu_get_be16(f); 2331 n_invalid = qemu_get_be16(f); 2332 2333 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) { 2334 /* End of Stream */ 2335 break; 2336 } 2337 2338 if ((index + n_valid + n_invalid) > 2339 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) { 2340 /* Bad index in stream */ 2341 error_report( 2342 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)", 2343 index, n_valid, n_invalid, spapr->htab_shift); 2344 return -EINVAL; 2345 } 2346 2347 if (spapr->htab) { 2348 if (n_valid) { 2349 qemu_get_buffer(f, HPTE(spapr->htab, index), 2350 HASH_PTE_SIZE_64 * n_valid); 2351 } 2352 if (n_invalid) { 2353 memset(HPTE(spapr->htab, index + n_valid), 0, 2354 HASH_PTE_SIZE_64 * n_invalid); 2355 } 2356 } else { 2357 int rc; 2358 2359 assert(fd >= 0); 2360 2361 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid, 2362 &local_err); 2363 if (rc < 0) { 2364 error_report_err(local_err); 2365 return rc; 2366 } 2367 } 2368 } 2369 2370 if (!spapr->htab) { 2371 assert(fd >= 0); 2372 close(fd); 2373 } 2374 2375 return 0; 2376 } 2377 2378 static void htab_save_cleanup(void *opaque) 2379 { 2380 SpaprMachineState *spapr = opaque; 2381 2382 close_htab_fd(spapr); 2383 } 2384 2385 static SaveVMHandlers savevm_htab_handlers = { 2386 .save_setup = htab_save_setup, 2387 .save_live_iterate = htab_save_iterate, 2388 .save_live_complete_precopy = htab_save_complete, 2389 .save_cleanup = htab_save_cleanup, 2390 .load_state = htab_load, 2391 }; 2392 2393 static void spapr_boot_set(void *opaque, const char *boot_device, 2394 Error **errp) 2395 { 2396 SpaprMachineState *spapr = SPAPR_MACHINE(opaque); 2397 2398 g_free(spapr->boot_device); 2399 spapr->boot_device = g_strdup(boot_device); 2400 } 2401 2402 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr) 2403 { 2404 MachineState *machine = MACHINE(spapr); 2405 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE; 2406 uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size; 2407 int i; 2408 2409 for (i = 0; i < nr_lmbs; i++) { 2410 uint64_t addr; 2411 2412 addr = i * lmb_size + machine->device_memory->base; 2413 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB, 2414 addr / lmb_size); 2415 } 2416 } 2417 2418 /* 2419 * If RAM size, maxmem size and individual node mem sizes aren't aligned 2420 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest 2421 * since we can't support such unaligned sizes with DRCONF_MEMORY. 2422 */ 2423 static void spapr_validate_node_memory(MachineState *machine, Error **errp) 2424 { 2425 int i; 2426 2427 if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) { 2428 error_setg(errp, "Memory size 0x" RAM_ADDR_FMT 2429 " is not aligned to %" PRIu64 " MiB", 2430 machine->ram_size, 2431 SPAPR_MEMORY_BLOCK_SIZE / MiB); 2432 return; 2433 } 2434 2435 if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) { 2436 error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT 2437 " is not aligned to %" PRIu64 " MiB", 2438 machine->ram_size, 2439 SPAPR_MEMORY_BLOCK_SIZE / MiB); 2440 return; 2441 } 2442 2443 for (i = 0; i < machine->numa_state->num_nodes; i++) { 2444 if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) { 2445 error_setg(errp, 2446 "Node %d memory size 0x%" PRIx64 2447 " is not aligned to %" PRIu64 " MiB", 2448 i, machine->numa_state->nodes[i].node_mem, 2449 SPAPR_MEMORY_BLOCK_SIZE / MiB); 2450 return; 2451 } 2452 } 2453 } 2454 2455 /* find cpu slot in machine->possible_cpus by core_id */ 2456 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx) 2457 { 2458 int index = id / ms->smp.threads; 2459 2460 if (index >= ms->possible_cpus->len) { 2461 return NULL; 2462 } 2463 if (idx) { 2464 *idx = index; 2465 } 2466 return &ms->possible_cpus->cpus[index]; 2467 } 2468 2469 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp) 2470 { 2471 MachineState *ms = MACHINE(spapr); 2472 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 2473 Error *local_err = NULL; 2474 bool vsmt_user = !!spapr->vsmt; 2475 int kvm_smt = kvmppc_smt_threads(); 2476 int ret; 2477 unsigned int smp_threads = ms->smp.threads; 2478 2479 if (!kvm_enabled() && (smp_threads > 1)) { 2480 error_setg(errp, "TCG cannot support more than 1 thread/core " 2481 "on a pseries machine"); 2482 return; 2483 } 2484 if (!is_power_of_2(smp_threads)) { 2485 error_setg(errp, "Cannot support %d threads/core on a pseries " 2486 "machine because it must be a power of 2", smp_threads); 2487 return; 2488 } 2489 2490 /* Detemine the VSMT mode to use: */ 2491 if (vsmt_user) { 2492 if (spapr->vsmt < smp_threads) { 2493 error_setg(errp, "Cannot support VSMT mode %d" 2494 " because it must be >= threads/core (%d)", 2495 spapr->vsmt, smp_threads); 2496 return; 2497 } 2498 /* In this case, spapr->vsmt has been set by the command line */ 2499 } else if (!smc->smp_threads_vsmt) { 2500 /* 2501 * Default VSMT value is tricky, because we need it to be as 2502 * consistent as possible (for migration), but this requires 2503 * changing it for at least some existing cases. We pick 8 as 2504 * the value that we'd get with KVM on POWER8, the 2505 * overwhelmingly common case in production systems. 2506 */ 2507 spapr->vsmt = MAX(8, smp_threads); 2508 } else { 2509 spapr->vsmt = smp_threads; 2510 } 2511 2512 /* KVM: If necessary, set the SMT mode: */ 2513 if (kvm_enabled() && (spapr->vsmt != kvm_smt)) { 2514 ret = kvmppc_set_smt_threads(spapr->vsmt); 2515 if (ret) { 2516 /* Looks like KVM isn't able to change VSMT mode */ 2517 error_setg(&local_err, 2518 "Failed to set KVM's VSMT mode to %d (errno %d)", 2519 spapr->vsmt, ret); 2520 /* We can live with that if the default one is big enough 2521 * for the number of threads, and a submultiple of the one 2522 * we want. In this case we'll waste some vcpu ids, but 2523 * behaviour will be correct */ 2524 if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) { 2525 warn_report_err(local_err); 2526 } else { 2527 if (!vsmt_user) { 2528 error_append_hint(&local_err, 2529 "On PPC, a VM with %d threads/core" 2530 " on a host with %d threads/core" 2531 " requires the use of VSMT mode %d.\n", 2532 smp_threads, kvm_smt, spapr->vsmt); 2533 } 2534 kvmppc_error_append_smt_possible_hint(&local_err); 2535 error_propagate(errp, local_err); 2536 } 2537 } 2538 } 2539 /* else TCG: nothing to do currently */ 2540 } 2541 2542 static void spapr_init_cpus(SpaprMachineState *spapr) 2543 { 2544 MachineState *machine = MACHINE(spapr); 2545 MachineClass *mc = MACHINE_GET_CLASS(machine); 2546 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine); 2547 const char *type = spapr_get_cpu_core_type(machine->cpu_type); 2548 const CPUArchIdList *possible_cpus; 2549 unsigned int smp_cpus = machine->smp.cpus; 2550 unsigned int smp_threads = machine->smp.threads; 2551 unsigned int max_cpus = machine->smp.max_cpus; 2552 int boot_cores_nr = smp_cpus / smp_threads; 2553 int i; 2554 2555 possible_cpus = mc->possible_cpu_arch_ids(machine); 2556 if (mc->has_hotpluggable_cpus) { 2557 if (smp_cpus % smp_threads) { 2558 error_report("smp_cpus (%u) must be multiple of threads (%u)", 2559 smp_cpus, smp_threads); 2560 exit(1); 2561 } 2562 if (max_cpus % smp_threads) { 2563 error_report("max_cpus (%u) must be multiple of threads (%u)", 2564 max_cpus, smp_threads); 2565 exit(1); 2566 } 2567 } else { 2568 if (max_cpus != smp_cpus) { 2569 error_report("This machine version does not support CPU hotplug"); 2570 exit(1); 2571 } 2572 boot_cores_nr = possible_cpus->len; 2573 } 2574 2575 if (smc->pre_2_10_has_unused_icps) { 2576 int i; 2577 2578 for (i = 0; i < spapr_max_server_number(spapr); i++) { 2579 /* Dummy entries get deregistered when real ICPState objects 2580 * are registered during CPU core hotplug. 2581 */ 2582 pre_2_10_vmstate_register_dummy_icp(i); 2583 } 2584 } 2585 2586 for (i = 0; i < possible_cpus->len; i++) { 2587 int core_id = i * smp_threads; 2588 2589 if (mc->has_hotpluggable_cpus) { 2590 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU, 2591 spapr_vcpu_id(spapr, core_id)); 2592 } 2593 2594 if (i < boot_cores_nr) { 2595 Object *core = object_new(type); 2596 int nr_threads = smp_threads; 2597 2598 /* Handle the partially filled core for older machine types */ 2599 if ((i + 1) * smp_threads >= smp_cpus) { 2600 nr_threads = smp_cpus - i * smp_threads; 2601 } 2602 2603 object_property_set_int(core, "nr-threads", nr_threads, 2604 &error_fatal); 2605 object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id, 2606 &error_fatal); 2607 qdev_realize(DEVICE(core), NULL, &error_fatal); 2608 2609 object_unref(core); 2610 } 2611 } 2612 } 2613 2614 static PCIHostState *spapr_create_default_phb(void) 2615 { 2616 DeviceState *dev; 2617 2618 dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE); 2619 qdev_prop_set_uint32(dev, "index", 0); 2620 sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal); 2621 2622 return PCI_HOST_BRIDGE(dev); 2623 } 2624 2625 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp) 2626 { 2627 MachineState *machine = MACHINE(spapr); 2628 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 2629 hwaddr rma_size = machine->ram_size; 2630 hwaddr node0_size = spapr_node0_size(machine); 2631 2632 /* RMA has to fit in the first NUMA node */ 2633 rma_size = MIN(rma_size, node0_size); 2634 2635 /* 2636 * VRMA access is via a special 1TiB SLB mapping, so the RMA can 2637 * never exceed that 2638 */ 2639 rma_size = MIN(rma_size, 1 * TiB); 2640 2641 /* 2642 * Clamp the RMA size based on machine type. This is for 2643 * migration compatibility with older qemu versions, which limited 2644 * the RMA size for complicated and mostly bad reasons. 2645 */ 2646 if (smc->rma_limit) { 2647 rma_size = MIN(rma_size, smc->rma_limit); 2648 } 2649 2650 if (rma_size < MIN_RMA_SLOF) { 2651 error_setg(errp, 2652 "pSeries SLOF firmware requires >= %" HWADDR_PRIx 2653 "ldMiB guest RMA (Real Mode Area memory)", 2654 MIN_RMA_SLOF / MiB); 2655 return 0; 2656 } 2657 2658 return rma_size; 2659 } 2660 2661 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr) 2662 { 2663 MachineState *machine = MACHINE(spapr); 2664 int i; 2665 2666 for (i = 0; i < machine->ram_slots; i++) { 2667 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i); 2668 } 2669 } 2670 2671 /* pSeries LPAR / sPAPR hardware init */ 2672 static void spapr_machine_init(MachineState *machine) 2673 { 2674 SpaprMachineState *spapr = SPAPR_MACHINE(machine); 2675 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine); 2676 MachineClass *mc = MACHINE_GET_CLASS(machine); 2677 const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME; 2678 const char *bios_name = machine->firmware ?: bios_default; 2679 const char *kernel_filename = machine->kernel_filename; 2680 const char *initrd_filename = machine->initrd_filename; 2681 PCIHostState *phb; 2682 int i; 2683 MemoryRegion *sysmem = get_system_memory(); 2684 long load_limit, fw_size; 2685 char *filename; 2686 Error *resize_hpt_err = NULL; 2687 2688 /* 2689 * if Secure VM (PEF) support is configured, then initialize it 2690 */ 2691 pef_kvm_init(machine->cgs, &error_fatal); 2692 2693 msi_nonbroken = true; 2694 2695 QLIST_INIT(&spapr->phbs); 2696 QTAILQ_INIT(&spapr->pending_dimm_unplugs); 2697 2698 /* Determine capabilities to run with */ 2699 spapr_caps_init(spapr); 2700 2701 kvmppc_check_papr_resize_hpt(&resize_hpt_err); 2702 if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) { 2703 /* 2704 * If the user explicitly requested a mode we should either 2705 * supply it, or fail completely (which we do below). But if 2706 * it's not set explicitly, we reset our mode to something 2707 * that works 2708 */ 2709 if (resize_hpt_err) { 2710 spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED; 2711 error_free(resize_hpt_err); 2712 resize_hpt_err = NULL; 2713 } else { 2714 spapr->resize_hpt = smc->resize_hpt_default; 2715 } 2716 } 2717 2718 assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT); 2719 2720 if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) { 2721 /* 2722 * User requested HPT resize, but this host can't supply it. Bail out 2723 */ 2724 error_report_err(resize_hpt_err); 2725 exit(1); 2726 } 2727 error_free(resize_hpt_err); 2728 2729 spapr->rma_size = spapr_rma_size(spapr, &error_fatal); 2730 2731 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */ 2732 load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD; 2733 2734 /* 2735 * VSMT must be set in order to be able to compute VCPU ids, ie to 2736 * call spapr_max_server_number() or spapr_vcpu_id(). 2737 */ 2738 spapr_set_vsmt_mode(spapr, &error_fatal); 2739 2740 /* Set up Interrupt Controller before we create the VCPUs */ 2741 spapr_irq_init(spapr, &error_fatal); 2742 2743 /* Set up containers for ibm,client-architecture-support negotiated options 2744 */ 2745 spapr->ov5 = spapr_ovec_new(); 2746 spapr->ov5_cas = spapr_ovec_new(); 2747 2748 if (smc->dr_lmb_enabled) { 2749 spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY); 2750 spapr_validate_node_memory(machine, &error_fatal); 2751 } 2752 2753 spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY); 2754 2755 /* advertise support for dedicated HP event source to guests */ 2756 if (spapr->use_hotplug_event_source) { 2757 spapr_ovec_set(spapr->ov5, OV5_HP_EVT); 2758 } 2759 2760 /* advertise support for HPT resizing */ 2761 if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) { 2762 spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE); 2763 } 2764 2765 /* advertise support for ibm,dyamic-memory-v2 */ 2766 spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2); 2767 2768 /* advertise XIVE on POWER9 machines */ 2769 if (spapr->irq->xive) { 2770 spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT); 2771 } 2772 2773 /* init CPUs */ 2774 spapr_init_cpus(spapr); 2775 2776 /* 2777 * check we don't have a memory-less/cpu-less NUMA node 2778 * Firmware relies on the existing memory/cpu topology to provide the 2779 * NUMA topology to the kernel. 2780 * And the linux kernel needs to know the NUMA topology at start 2781 * to be able to hotplug CPUs later. 2782 */ 2783 if (machine->numa_state->num_nodes) { 2784 for (i = 0; i < machine->numa_state->num_nodes; ++i) { 2785 /* check for memory-less node */ 2786 if (machine->numa_state->nodes[i].node_mem == 0) { 2787 CPUState *cs; 2788 int found = 0; 2789 /* check for cpu-less node */ 2790 CPU_FOREACH(cs) { 2791 PowerPCCPU *cpu = POWERPC_CPU(cs); 2792 if (cpu->node_id == i) { 2793 found = 1; 2794 break; 2795 } 2796 } 2797 /* memory-less and cpu-less node */ 2798 if (!found) { 2799 error_report( 2800 "Memory-less/cpu-less nodes are not supported (node %d)", 2801 i); 2802 exit(1); 2803 } 2804 } 2805 } 2806 2807 } 2808 2809 spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine); 2810 2811 /* Init numa_assoc_array */ 2812 spapr_numa_associativity_init(spapr, machine); 2813 2814 if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) && 2815 ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0, 2816 spapr->max_compat_pvr)) { 2817 spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300); 2818 /* KVM and TCG always allow GTSE with radix... */ 2819 spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE); 2820 } 2821 /* ... but not with hash (currently). */ 2822 2823 if (kvm_enabled()) { 2824 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */ 2825 kvmppc_enable_logical_ci_hcalls(); 2826 kvmppc_enable_set_mode_hcall(); 2827 2828 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */ 2829 kvmppc_enable_clear_ref_mod_hcalls(); 2830 2831 /* Enable H_PAGE_INIT */ 2832 kvmppc_enable_h_page_init(); 2833 } 2834 2835 /* map RAM */ 2836 memory_region_add_subregion(sysmem, 0, machine->ram); 2837 2838 /* always allocate the device memory information */ 2839 machine->device_memory = g_malloc0(sizeof(*machine->device_memory)); 2840 2841 /* initialize hotplug memory address space */ 2842 if (machine->ram_size < machine->maxram_size) { 2843 ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size; 2844 /* 2845 * Limit the number of hotpluggable memory slots to half the number 2846 * slots that KVM supports, leaving the other half for PCI and other 2847 * devices. However ensure that number of slots doesn't drop below 32. 2848 */ 2849 int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 : 2850 SPAPR_MAX_RAM_SLOTS; 2851 2852 if (max_memslots < SPAPR_MAX_RAM_SLOTS) { 2853 max_memslots = SPAPR_MAX_RAM_SLOTS; 2854 } 2855 if (machine->ram_slots > max_memslots) { 2856 error_report("Specified number of memory slots %" 2857 PRIu64" exceeds max supported %d", 2858 machine->ram_slots, max_memslots); 2859 exit(1); 2860 } 2861 2862 machine->device_memory->base = ROUND_UP(machine->ram_size, 2863 SPAPR_DEVICE_MEM_ALIGN); 2864 memory_region_init(&machine->device_memory->mr, OBJECT(spapr), 2865 "device-memory", device_mem_size); 2866 memory_region_add_subregion(sysmem, machine->device_memory->base, 2867 &machine->device_memory->mr); 2868 } 2869 2870 if (smc->dr_lmb_enabled) { 2871 spapr_create_lmb_dr_connectors(spapr); 2872 } 2873 2874 if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) { 2875 /* Create the error string for live migration blocker */ 2876 error_setg(&spapr->fwnmi_migration_blocker, 2877 "A machine check is being handled during migration. The handler" 2878 "may run and log hardware error on the destination"); 2879 } 2880 2881 if (mc->nvdimm_supported) { 2882 spapr_create_nvdimm_dr_connectors(spapr); 2883 } 2884 2885 /* Set up RTAS event infrastructure */ 2886 spapr_events_init(spapr); 2887 2888 /* Set up the RTC RTAS interfaces */ 2889 spapr_rtc_create(spapr); 2890 2891 /* Set up VIO bus */ 2892 spapr->vio_bus = spapr_vio_bus_init(); 2893 2894 for (i = 0; serial_hd(i); i++) { 2895 spapr_vty_create(spapr->vio_bus, serial_hd(i)); 2896 } 2897 2898 /* We always have at least the nvram device on VIO */ 2899 spapr_create_nvram(spapr); 2900 2901 /* 2902 * Setup hotplug / dynamic-reconfiguration connectors. top-level 2903 * connectors (described in root DT node's "ibm,drc-types" property) 2904 * are pre-initialized here. additional child connectors (such as 2905 * connectors for a PHBs PCI slots) are added as needed during their 2906 * parent's realization. 2907 */ 2908 if (smc->dr_phb_enabled) { 2909 for (i = 0; i < SPAPR_MAX_PHBS; i++) { 2910 spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i); 2911 } 2912 } 2913 2914 /* Set up PCI */ 2915 spapr_pci_rtas_init(); 2916 2917 phb = spapr_create_default_phb(); 2918 2919 for (i = 0; i < nb_nics; i++) { 2920 NICInfo *nd = &nd_table[i]; 2921 2922 if (!nd->model) { 2923 nd->model = g_strdup("spapr-vlan"); 2924 } 2925 2926 if (g_str_equal(nd->model, "spapr-vlan") || 2927 g_str_equal(nd->model, "ibmveth")) { 2928 spapr_vlan_create(spapr->vio_bus, nd); 2929 } else { 2930 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL); 2931 } 2932 } 2933 2934 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) { 2935 spapr_vscsi_create(spapr->vio_bus); 2936 } 2937 2938 /* Graphics */ 2939 if (spapr_vga_init(phb->bus, &error_fatal)) { 2940 spapr->has_graphics = true; 2941 machine->usb |= defaults_enabled() && !machine->usb_disabled; 2942 } 2943 2944 if (machine->usb) { 2945 if (smc->use_ohci_by_default) { 2946 pci_create_simple(phb->bus, -1, "pci-ohci"); 2947 } else { 2948 pci_create_simple(phb->bus, -1, "nec-usb-xhci"); 2949 } 2950 2951 if (spapr->has_graphics) { 2952 USBBus *usb_bus = usb_bus_find(-1); 2953 2954 usb_create_simple(usb_bus, "usb-kbd"); 2955 usb_create_simple(usb_bus, "usb-mouse"); 2956 } 2957 } 2958 2959 if (kernel_filename) { 2960 spapr->kernel_size = load_elf(kernel_filename, NULL, 2961 translate_kernel_address, spapr, 2962 NULL, NULL, NULL, NULL, 1, 2963 PPC_ELF_MACHINE, 0, 0); 2964 if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) { 2965 spapr->kernel_size = load_elf(kernel_filename, NULL, 2966 translate_kernel_address, spapr, 2967 NULL, NULL, NULL, NULL, 0, 2968 PPC_ELF_MACHINE, 0, 0); 2969 spapr->kernel_le = spapr->kernel_size > 0; 2970 } 2971 if (spapr->kernel_size < 0) { 2972 error_report("error loading %s: %s", kernel_filename, 2973 load_elf_strerror(spapr->kernel_size)); 2974 exit(1); 2975 } 2976 2977 /* load initrd */ 2978 if (initrd_filename) { 2979 /* Try to locate the initrd in the gap between the kernel 2980 * and the firmware. Add a bit of space just in case 2981 */ 2982 spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size 2983 + 0x1ffff) & ~0xffff; 2984 spapr->initrd_size = load_image_targphys(initrd_filename, 2985 spapr->initrd_base, 2986 load_limit 2987 - spapr->initrd_base); 2988 if (spapr->initrd_size < 0) { 2989 error_report("could not load initial ram disk '%s'", 2990 initrd_filename); 2991 exit(1); 2992 } 2993 } 2994 } 2995 2996 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name); 2997 if (!filename) { 2998 error_report("Could not find LPAR firmware '%s'", bios_name); 2999 exit(1); 3000 } 3001 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE); 3002 if (fw_size <= 0) { 3003 error_report("Could not load LPAR firmware '%s'", filename); 3004 exit(1); 3005 } 3006 g_free(filename); 3007 3008 /* FIXME: Should register things through the MachineState's qdev 3009 * interface, this is a legacy from the sPAPREnvironment structure 3010 * which predated MachineState but had a similar function */ 3011 vmstate_register(NULL, 0, &vmstate_spapr, spapr); 3012 register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1, 3013 &savevm_htab_handlers, spapr); 3014 3015 qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine)); 3016 3017 qemu_register_boot_set(spapr_boot_set, spapr); 3018 3019 /* 3020 * Nothing needs to be done to resume a suspended guest because 3021 * suspending does not change the machine state, so no need for 3022 * a ->wakeup method. 3023 */ 3024 qemu_register_wakeup_support(); 3025 3026 if (kvm_enabled()) { 3027 /* to stop and start vmclock */ 3028 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change, 3029 &spapr->tb); 3030 3031 kvmppc_spapr_enable_inkernel_multitce(); 3032 } 3033 3034 qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond); 3035 if (spapr->vof) { 3036 spapr->vof->fw_size = fw_size; /* for claim() on itself */ 3037 spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client); 3038 } 3039 } 3040 3041 #define DEFAULT_KVM_TYPE "auto" 3042 static int spapr_kvm_type(MachineState *machine, const char *vm_type) 3043 { 3044 /* 3045 * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to 3046 * accomodate the 'HV' and 'PV' formats that exists in the 3047 * wild. The 'auto' mode is being introduced already as 3048 * lower-case, thus we don't need to bother checking for 3049 * "AUTO". 3050 */ 3051 if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) { 3052 return 0; 3053 } 3054 3055 if (!g_ascii_strcasecmp(vm_type, "hv")) { 3056 return 1; 3057 } 3058 3059 if (!g_ascii_strcasecmp(vm_type, "pr")) { 3060 return 2; 3061 } 3062 3063 error_report("Unknown kvm-type specified '%s'", vm_type); 3064 exit(1); 3065 } 3066 3067 /* 3068 * Implementation of an interface to adjust firmware path 3069 * for the bootindex property handling. 3070 */ 3071 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus, 3072 DeviceState *dev) 3073 { 3074 #define CAST(type, obj, name) \ 3075 ((type *)object_dynamic_cast(OBJECT(obj), (name))) 3076 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE); 3077 SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE); 3078 VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON); 3079 PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE); 3080 3081 if (d) { 3082 void *spapr = CAST(void, bus->parent, "spapr-vscsi"); 3083 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI); 3084 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE); 3085 3086 if (spapr) { 3087 /* 3088 * Replace "channel@0/disk@0,0" with "disk@8000000000000000": 3089 * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form 3090 * 0x8000 | (target << 8) | (bus << 5) | lun 3091 * (see the "Logical unit addressing format" table in SAM5) 3092 */ 3093 unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun; 3094 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev), 3095 (uint64_t)id << 48); 3096 } else if (virtio) { 3097 /* 3098 * We use SRP luns of the form 01000000 | (target << 8) | lun 3099 * in the top 32 bits of the 64-bit LUN 3100 * Note: the quote above is from SLOF and it is wrong, 3101 * the actual binding is: 3102 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun ) 3103 */ 3104 unsigned id = 0x1000000 | (d->id << 16) | d->lun; 3105 if (d->lun >= 256) { 3106 /* Use the LUN "flat space addressing method" */ 3107 id |= 0x4000; 3108 } 3109 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev), 3110 (uint64_t)id << 32); 3111 } else if (usb) { 3112 /* 3113 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun 3114 * in the top 32 bits of the 64-bit LUN 3115 */ 3116 unsigned usb_port = atoi(usb->port->path); 3117 unsigned id = 0x1000000 | (usb_port << 16) | d->lun; 3118 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev), 3119 (uint64_t)id << 32); 3120 } 3121 } 3122 3123 /* 3124 * SLOF probes the USB devices, and if it recognizes that the device is a 3125 * storage device, it changes its name to "storage" instead of "usb-host", 3126 * and additionally adds a child node for the SCSI LUN, so the correct 3127 * boot path in SLOF is something like .../storage@1/disk@xxx" instead. 3128 */ 3129 if (strcmp("usb-host", qdev_fw_name(dev)) == 0) { 3130 USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE); 3131 if (usb_device_is_scsi_storage(usbdev)) { 3132 return g_strdup_printf("storage@%s/disk", usbdev->port->path); 3133 } 3134 } 3135 3136 if (phb) { 3137 /* Replace "pci" with "pci@800000020000000" */ 3138 return g_strdup_printf("pci@%"PRIX64, phb->buid); 3139 } 3140 3141 if (vsc) { 3142 /* Same logic as virtio above */ 3143 unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun; 3144 return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32); 3145 } 3146 3147 if (g_str_equal("pci-bridge", qdev_fw_name(dev))) { 3148 /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */ 3149 PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE); 3150 return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn)); 3151 } 3152 3153 if (pcidev) { 3154 return spapr_pci_fw_dev_name(pcidev); 3155 } 3156 3157 return NULL; 3158 } 3159 3160 static char *spapr_get_kvm_type(Object *obj, Error **errp) 3161 { 3162 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3163 3164 return g_strdup(spapr->kvm_type); 3165 } 3166 3167 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp) 3168 { 3169 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3170 3171 g_free(spapr->kvm_type); 3172 spapr->kvm_type = g_strdup(value); 3173 } 3174 3175 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp) 3176 { 3177 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3178 3179 return spapr->use_hotplug_event_source; 3180 } 3181 3182 static void spapr_set_modern_hotplug_events(Object *obj, bool value, 3183 Error **errp) 3184 { 3185 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3186 3187 spapr->use_hotplug_event_source = value; 3188 } 3189 3190 static bool spapr_get_msix_emulation(Object *obj, Error **errp) 3191 { 3192 return true; 3193 } 3194 3195 static char *spapr_get_resize_hpt(Object *obj, Error **errp) 3196 { 3197 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3198 3199 switch (spapr->resize_hpt) { 3200 case SPAPR_RESIZE_HPT_DEFAULT: 3201 return g_strdup("default"); 3202 case SPAPR_RESIZE_HPT_DISABLED: 3203 return g_strdup("disabled"); 3204 case SPAPR_RESIZE_HPT_ENABLED: 3205 return g_strdup("enabled"); 3206 case SPAPR_RESIZE_HPT_REQUIRED: 3207 return g_strdup("required"); 3208 } 3209 g_assert_not_reached(); 3210 } 3211 3212 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp) 3213 { 3214 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3215 3216 if (strcmp(value, "default") == 0) { 3217 spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT; 3218 } else if (strcmp(value, "disabled") == 0) { 3219 spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED; 3220 } else if (strcmp(value, "enabled") == 0) { 3221 spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED; 3222 } else if (strcmp(value, "required") == 0) { 3223 spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED; 3224 } else { 3225 error_setg(errp, "Bad value for \"resize-hpt\" property"); 3226 } 3227 } 3228 3229 static bool spapr_get_vof(Object *obj, Error **errp) 3230 { 3231 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3232 3233 return spapr->vof != NULL; 3234 } 3235 3236 static void spapr_set_vof(Object *obj, bool value, Error **errp) 3237 { 3238 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3239 3240 if (spapr->vof) { 3241 vof_cleanup(spapr->vof); 3242 g_free(spapr->vof); 3243 spapr->vof = NULL; 3244 } 3245 if (!value) { 3246 return; 3247 } 3248 spapr->vof = g_malloc0(sizeof(*spapr->vof)); 3249 } 3250 3251 static char *spapr_get_ic_mode(Object *obj, Error **errp) 3252 { 3253 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3254 3255 if (spapr->irq == &spapr_irq_xics_legacy) { 3256 return g_strdup("legacy"); 3257 } else if (spapr->irq == &spapr_irq_xics) { 3258 return g_strdup("xics"); 3259 } else if (spapr->irq == &spapr_irq_xive) { 3260 return g_strdup("xive"); 3261 } else if (spapr->irq == &spapr_irq_dual) { 3262 return g_strdup("dual"); 3263 } 3264 g_assert_not_reached(); 3265 } 3266 3267 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp) 3268 { 3269 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3270 3271 if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) { 3272 error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode"); 3273 return; 3274 } 3275 3276 /* The legacy IRQ backend can not be set */ 3277 if (strcmp(value, "xics") == 0) { 3278 spapr->irq = &spapr_irq_xics; 3279 } else if (strcmp(value, "xive") == 0) { 3280 spapr->irq = &spapr_irq_xive; 3281 } else if (strcmp(value, "dual") == 0) { 3282 spapr->irq = &spapr_irq_dual; 3283 } else { 3284 error_setg(errp, "Bad value for \"ic-mode\" property"); 3285 } 3286 } 3287 3288 static char *spapr_get_host_model(Object *obj, Error **errp) 3289 { 3290 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3291 3292 return g_strdup(spapr->host_model); 3293 } 3294 3295 static void spapr_set_host_model(Object *obj, const char *value, Error **errp) 3296 { 3297 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3298 3299 g_free(spapr->host_model); 3300 spapr->host_model = g_strdup(value); 3301 } 3302 3303 static char *spapr_get_host_serial(Object *obj, Error **errp) 3304 { 3305 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3306 3307 return g_strdup(spapr->host_serial); 3308 } 3309 3310 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp) 3311 { 3312 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3313 3314 g_free(spapr->host_serial); 3315 spapr->host_serial = g_strdup(value); 3316 } 3317 3318 static void spapr_instance_init(Object *obj) 3319 { 3320 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3321 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 3322 MachineState *ms = MACHINE(spapr); 3323 MachineClass *mc = MACHINE_GET_CLASS(ms); 3324 3325 /* 3326 * NVDIMM support went live in 5.1 without considering that, in 3327 * other archs, the user needs to enable NVDIMM support with the 3328 * 'nvdimm' machine option and the default behavior is NVDIMM 3329 * support disabled. It is too late to roll back to the standard 3330 * behavior without breaking 5.1 guests. 3331 */ 3332 if (mc->nvdimm_supported) { 3333 ms->nvdimms_state->is_enabled = true; 3334 } 3335 3336 spapr->htab_fd = -1; 3337 spapr->use_hotplug_event_source = true; 3338 spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE); 3339 object_property_add_str(obj, "kvm-type", 3340 spapr_get_kvm_type, spapr_set_kvm_type); 3341 object_property_set_description(obj, "kvm-type", 3342 "Specifies the KVM virtualization mode (auto," 3343 " hv, pr). Defaults to 'auto'. This mode will use" 3344 " any available KVM module loaded in the host," 3345 " where kvm_hv takes precedence if both kvm_hv and" 3346 " kvm_pr are loaded."); 3347 object_property_add_bool(obj, "modern-hotplug-events", 3348 spapr_get_modern_hotplug_events, 3349 spapr_set_modern_hotplug_events); 3350 object_property_set_description(obj, "modern-hotplug-events", 3351 "Use dedicated hotplug event mechanism in" 3352 " place of standard EPOW events when possible" 3353 " (required for memory hot-unplug support)"); 3354 ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr, 3355 "Maximum permitted CPU compatibility mode"); 3356 3357 object_property_add_str(obj, "resize-hpt", 3358 spapr_get_resize_hpt, spapr_set_resize_hpt); 3359 object_property_set_description(obj, "resize-hpt", 3360 "Resizing of the Hash Page Table (enabled, disabled, required)"); 3361 object_property_add_uint32_ptr(obj, "vsmt", 3362 &spapr->vsmt, OBJ_PROP_FLAG_READWRITE); 3363 object_property_set_description(obj, "vsmt", 3364 "Virtual SMT: KVM behaves as if this were" 3365 " the host's SMT mode"); 3366 3367 object_property_add_bool(obj, "vfio-no-msix-emulation", 3368 spapr_get_msix_emulation, NULL); 3369 3370 object_property_add_uint64_ptr(obj, "kernel-addr", 3371 &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE); 3372 object_property_set_description(obj, "kernel-addr", 3373 stringify(KERNEL_LOAD_ADDR) 3374 " for -kernel is the default"); 3375 spapr->kernel_addr = KERNEL_LOAD_ADDR; 3376 3377 object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof); 3378 object_property_set_description(obj, "x-vof", 3379 "Enable Virtual Open Firmware (experimental)"); 3380 3381 /* The machine class defines the default interrupt controller mode */ 3382 spapr->irq = smc->irq; 3383 object_property_add_str(obj, "ic-mode", spapr_get_ic_mode, 3384 spapr_set_ic_mode); 3385 object_property_set_description(obj, "ic-mode", 3386 "Specifies the interrupt controller mode (xics, xive, dual)"); 3387 3388 object_property_add_str(obj, "host-model", 3389 spapr_get_host_model, spapr_set_host_model); 3390 object_property_set_description(obj, "host-model", 3391 "Host model to advertise in guest device tree"); 3392 object_property_add_str(obj, "host-serial", 3393 spapr_get_host_serial, spapr_set_host_serial); 3394 object_property_set_description(obj, "host-serial", 3395 "Host serial number to advertise in guest device tree"); 3396 } 3397 3398 static void spapr_machine_finalizefn(Object *obj) 3399 { 3400 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 3401 3402 g_free(spapr->kvm_type); 3403 } 3404 3405 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg) 3406 { 3407 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 3408 PowerPCCPU *cpu = POWERPC_CPU(cs); 3409 CPUPPCState *env = &cpu->env; 3410 3411 cpu_synchronize_state(cs); 3412 /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */ 3413 if (spapr->fwnmi_system_reset_addr != -1) { 3414 uint64_t rtas_addr, addr; 3415 3416 /* get rtas addr from fdt */ 3417 rtas_addr = spapr_get_rtas_addr(); 3418 if (!rtas_addr) { 3419 qemu_system_guest_panicked(NULL); 3420 return; 3421 } 3422 3423 addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2; 3424 stq_be_phys(&address_space_memory, addr, env->gpr[3]); 3425 stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0); 3426 env->gpr[3] = addr; 3427 } 3428 ppc_cpu_do_system_reset(cs); 3429 if (spapr->fwnmi_system_reset_addr != -1) { 3430 env->nip = spapr->fwnmi_system_reset_addr; 3431 } 3432 } 3433 3434 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp) 3435 { 3436 CPUState *cs; 3437 3438 CPU_FOREACH(cs) { 3439 async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL); 3440 } 3441 } 3442 3443 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr, 3444 void *fdt, int *fdt_start_offset, Error **errp) 3445 { 3446 uint64_t addr; 3447 uint32_t node; 3448 3449 addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE; 3450 node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP, 3451 &error_abort); 3452 *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr, 3453 SPAPR_MEMORY_BLOCK_SIZE); 3454 return 0; 3455 } 3456 3457 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size, 3458 bool dedicated_hp_event_source) 3459 { 3460 SpaprDrc *drc; 3461 uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE; 3462 int i; 3463 uint64_t addr = addr_start; 3464 bool hotplugged = spapr_drc_hotplugged(dev); 3465 3466 for (i = 0; i < nr_lmbs; i++) { 3467 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, 3468 addr / SPAPR_MEMORY_BLOCK_SIZE); 3469 g_assert(drc); 3470 3471 /* 3472 * memory_device_get_free_addr() provided a range of free addresses 3473 * that doesn't overlap with any existing mapping at pre-plug. The 3474 * corresponding LMB DRCs are thus assumed to be all attachable. 3475 */ 3476 spapr_drc_attach(drc, dev); 3477 if (!hotplugged) { 3478 spapr_drc_reset(drc); 3479 } 3480 addr += SPAPR_MEMORY_BLOCK_SIZE; 3481 } 3482 /* send hotplug notification to the 3483 * guest only in case of hotplugged memory 3484 */ 3485 if (hotplugged) { 3486 if (dedicated_hp_event_source) { 3487 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, 3488 addr_start / SPAPR_MEMORY_BLOCK_SIZE); 3489 g_assert(drc); 3490 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB, 3491 nr_lmbs, 3492 spapr_drc_index(drc)); 3493 } else { 3494 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB, 3495 nr_lmbs); 3496 } 3497 } 3498 } 3499 3500 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev) 3501 { 3502 SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev); 3503 PCDIMMDevice *dimm = PC_DIMM(dev); 3504 uint64_t size, addr; 3505 int64_t slot; 3506 bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM); 3507 3508 size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort); 3509 3510 pc_dimm_plug(dimm, MACHINE(ms)); 3511 3512 if (!is_nvdimm) { 3513 addr = object_property_get_uint(OBJECT(dimm), 3514 PC_DIMM_ADDR_PROP, &error_abort); 3515 spapr_add_lmbs(dev, addr, size, 3516 spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT)); 3517 } else { 3518 slot = object_property_get_int(OBJECT(dimm), 3519 PC_DIMM_SLOT_PROP, &error_abort); 3520 /* We should have valid slot number at this point */ 3521 g_assert(slot >= 0); 3522 spapr_add_nvdimm(dev, slot); 3523 } 3524 } 3525 3526 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, 3527 Error **errp) 3528 { 3529 const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev); 3530 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev); 3531 bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM); 3532 PCDIMMDevice *dimm = PC_DIMM(dev); 3533 Error *local_err = NULL; 3534 uint64_t size; 3535 Object *memdev; 3536 hwaddr pagesize; 3537 3538 if (!smc->dr_lmb_enabled) { 3539 error_setg(errp, "Memory hotplug not supported for this machine"); 3540 return; 3541 } 3542 3543 size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err); 3544 if (local_err) { 3545 error_propagate(errp, local_err); 3546 return; 3547 } 3548 3549 if (is_nvdimm) { 3550 if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) { 3551 return; 3552 } 3553 } else if (size % SPAPR_MEMORY_BLOCK_SIZE) { 3554 error_setg(errp, "Hotplugged memory size must be a multiple of " 3555 "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB); 3556 return; 3557 } 3558 3559 memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, 3560 &error_abort); 3561 pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev)); 3562 if (!spapr_check_pagesize(spapr, pagesize, errp)) { 3563 return; 3564 } 3565 3566 pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp); 3567 } 3568 3569 struct SpaprDimmState { 3570 PCDIMMDevice *dimm; 3571 uint32_t nr_lmbs; 3572 QTAILQ_ENTRY(SpaprDimmState) next; 3573 }; 3574 3575 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s, 3576 PCDIMMDevice *dimm) 3577 { 3578 SpaprDimmState *dimm_state = NULL; 3579 3580 QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) { 3581 if (dimm_state->dimm == dimm) { 3582 break; 3583 } 3584 } 3585 return dimm_state; 3586 } 3587 3588 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr, 3589 uint32_t nr_lmbs, 3590 PCDIMMDevice *dimm) 3591 { 3592 SpaprDimmState *ds = NULL; 3593 3594 /* 3595 * If this request is for a DIMM whose removal had failed earlier 3596 * (due to guest's refusal to remove the LMBs), we would have this 3597 * dimm already in the pending_dimm_unplugs list. In that 3598 * case don't add again. 3599 */ 3600 ds = spapr_pending_dimm_unplugs_find(spapr, dimm); 3601 if (!ds) { 3602 ds = g_malloc0(sizeof(SpaprDimmState)); 3603 ds->nr_lmbs = nr_lmbs; 3604 ds->dimm = dimm; 3605 QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next); 3606 } 3607 return ds; 3608 } 3609 3610 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr, 3611 SpaprDimmState *dimm_state) 3612 { 3613 QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next); 3614 g_free(dimm_state); 3615 } 3616 3617 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms, 3618 PCDIMMDevice *dimm) 3619 { 3620 SpaprDrc *drc; 3621 uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm), 3622 &error_abort); 3623 uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE; 3624 uint32_t avail_lmbs = 0; 3625 uint64_t addr_start, addr; 3626 int i; 3627 3628 addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP, 3629 &error_abort); 3630 3631 addr = addr_start; 3632 for (i = 0; i < nr_lmbs; i++) { 3633 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, 3634 addr / SPAPR_MEMORY_BLOCK_SIZE); 3635 g_assert(drc); 3636 if (drc->dev) { 3637 avail_lmbs++; 3638 } 3639 addr += SPAPR_MEMORY_BLOCK_SIZE; 3640 } 3641 3642 return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm); 3643 } 3644 3645 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev) 3646 { 3647 SpaprDimmState *ds; 3648 PCDIMMDevice *dimm; 3649 SpaprDrc *drc; 3650 uint32_t nr_lmbs; 3651 uint64_t size, addr_start, addr; 3652 g_autofree char *qapi_error = NULL; 3653 int i; 3654 3655 if (!dev) { 3656 return; 3657 } 3658 3659 dimm = PC_DIMM(dev); 3660 ds = spapr_pending_dimm_unplugs_find(spapr, dimm); 3661 3662 /* 3663 * 'ds == NULL' would mean that the DIMM doesn't have a pending 3664 * unplug state, but one of its DRC is marked as unplug_requested. 3665 * This is bad and weird enough to g_assert() out. 3666 */ 3667 g_assert(ds); 3668 3669 spapr_pending_dimm_unplugs_remove(spapr, ds); 3670 3671 size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort); 3672 nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE; 3673 3674 addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP, 3675 &error_abort); 3676 3677 addr = addr_start; 3678 for (i = 0; i < nr_lmbs; i++) { 3679 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, 3680 addr / SPAPR_MEMORY_BLOCK_SIZE); 3681 g_assert(drc); 3682 3683 drc->unplug_requested = false; 3684 addr += SPAPR_MEMORY_BLOCK_SIZE; 3685 } 3686 3687 /* 3688 * Tell QAPI that something happened and the memory 3689 * hotunplug wasn't successful. 3690 */ 3691 qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest " 3692 "for device %s", dev->id); 3693 qapi_event_send_mem_unplug_error(dev->id, qapi_error); 3694 } 3695 3696 /* Callback to be called during DRC release. */ 3697 void spapr_lmb_release(DeviceState *dev) 3698 { 3699 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev); 3700 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl); 3701 SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev)); 3702 3703 /* This information will get lost if a migration occurs 3704 * during the unplug process. In this case recover it. */ 3705 if (ds == NULL) { 3706 ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev)); 3707 g_assert(ds); 3708 /* The DRC being examined by the caller at least must be counted */ 3709 g_assert(ds->nr_lmbs); 3710 } 3711 3712 if (--ds->nr_lmbs) { 3713 return; 3714 } 3715 3716 /* 3717 * Now that all the LMBs have been removed by the guest, call the 3718 * unplug handler chain. This can never fail. 3719 */ 3720 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort); 3721 object_unparent(OBJECT(dev)); 3722 } 3723 3724 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev) 3725 { 3726 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev); 3727 SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev)); 3728 3729 /* We really shouldn't get this far without anything to unplug */ 3730 g_assert(ds); 3731 3732 pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev)); 3733 qdev_unrealize(dev); 3734 spapr_pending_dimm_unplugs_remove(spapr, ds); 3735 } 3736 3737 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev, 3738 DeviceState *dev, Error **errp) 3739 { 3740 SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev); 3741 PCDIMMDevice *dimm = PC_DIMM(dev); 3742 uint32_t nr_lmbs; 3743 uint64_t size, addr_start, addr; 3744 int i; 3745 SpaprDrc *drc; 3746 3747 if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) { 3748 error_setg(errp, "nvdimm device hot unplug is not supported yet."); 3749 return; 3750 } 3751 3752 size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort); 3753 nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE; 3754 3755 addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP, 3756 &error_abort); 3757 3758 /* 3759 * An existing pending dimm state for this DIMM means that there is an 3760 * unplug operation in progress, waiting for the spapr_lmb_release 3761 * callback to complete the job (BQL can't cover that far). In this case, 3762 * bail out to avoid detaching DRCs that were already released. 3763 */ 3764 if (spapr_pending_dimm_unplugs_find(spapr, dimm)) { 3765 error_setg(errp, "Memory unplug already in progress for device %s", 3766 dev->id); 3767 return; 3768 } 3769 3770 spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm); 3771 3772 addr = addr_start; 3773 for (i = 0; i < nr_lmbs; i++) { 3774 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, 3775 addr / SPAPR_MEMORY_BLOCK_SIZE); 3776 g_assert(drc); 3777 3778 spapr_drc_unplug_request(drc); 3779 addr += SPAPR_MEMORY_BLOCK_SIZE; 3780 } 3781 3782 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, 3783 addr_start / SPAPR_MEMORY_BLOCK_SIZE); 3784 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB, 3785 nr_lmbs, spapr_drc_index(drc)); 3786 } 3787 3788 /* Callback to be called during DRC release. */ 3789 void spapr_core_release(DeviceState *dev) 3790 { 3791 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev); 3792 3793 /* Call the unplug handler chain. This can never fail. */ 3794 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort); 3795 object_unparent(OBJECT(dev)); 3796 } 3797 3798 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev) 3799 { 3800 MachineState *ms = MACHINE(hotplug_dev); 3801 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms); 3802 CPUCore *cc = CPU_CORE(dev); 3803 CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL); 3804 3805 if (smc->pre_2_10_has_unused_icps) { 3806 SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev)); 3807 int i; 3808 3809 for (i = 0; i < cc->nr_threads; i++) { 3810 CPUState *cs = CPU(sc->threads[i]); 3811 3812 pre_2_10_vmstate_register_dummy_icp(cs->cpu_index); 3813 } 3814 } 3815 3816 assert(core_slot); 3817 core_slot->cpu = NULL; 3818 qdev_unrealize(dev); 3819 } 3820 3821 static 3822 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev, 3823 Error **errp) 3824 { 3825 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev)); 3826 int index; 3827 SpaprDrc *drc; 3828 CPUCore *cc = CPU_CORE(dev); 3829 3830 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) { 3831 error_setg(errp, "Unable to find CPU core with core-id: %d", 3832 cc->core_id); 3833 return; 3834 } 3835 if (index == 0) { 3836 error_setg(errp, "Boot CPU core may not be unplugged"); 3837 return; 3838 } 3839 3840 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, 3841 spapr_vcpu_id(spapr, cc->core_id)); 3842 g_assert(drc); 3843 3844 if (!spapr_drc_unplug_requested(drc)) { 3845 spapr_drc_unplug_request(drc); 3846 } 3847 3848 /* 3849 * spapr_hotplug_req_remove_by_index is left unguarded, out of the 3850 * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ 3851 * pulses removing the same CPU. Otherwise, in an failed hotunplug 3852 * attempt (e.g. the kernel will refuse to remove the last online 3853 * CPU), we will never attempt it again because unplug_requested 3854 * will still be 'true' in that case. 3855 */ 3856 spapr_hotplug_req_remove_by_index(drc); 3857 } 3858 3859 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr, 3860 void *fdt, int *fdt_start_offset, Error **errp) 3861 { 3862 SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev); 3863 CPUState *cs = CPU(core->threads[0]); 3864 PowerPCCPU *cpu = POWERPC_CPU(cs); 3865 DeviceClass *dc = DEVICE_GET_CLASS(cs); 3866 int id = spapr_get_vcpu_id(cpu); 3867 g_autofree char *nodename = NULL; 3868 int offset; 3869 3870 nodename = g_strdup_printf("%s@%x", dc->fw_name, id); 3871 offset = fdt_add_subnode(fdt, 0, nodename); 3872 3873 spapr_dt_cpu(cs, fdt, offset, spapr); 3874 3875 /* 3876 * spapr_dt_cpu() does not fill the 'name' property in the 3877 * CPU node. The function is called during boot process, before 3878 * and after CAS, and overwriting the 'name' property written 3879 * by SLOF is not allowed. 3880 * 3881 * Write it manually after spapr_dt_cpu(). This makes the hotplug 3882 * CPUs more compatible with the coldplugged ones, which have 3883 * the 'name' property. Linux Kernel also relies on this 3884 * property to identify CPU nodes. 3885 */ 3886 _FDT((fdt_setprop_string(fdt, offset, "name", nodename))); 3887 3888 *fdt_start_offset = offset; 3889 return 0; 3890 } 3891 3892 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev) 3893 { 3894 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev)); 3895 MachineClass *mc = MACHINE_GET_CLASS(spapr); 3896 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 3897 SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev)); 3898 CPUCore *cc = CPU_CORE(dev); 3899 CPUState *cs; 3900 SpaprDrc *drc; 3901 CPUArchId *core_slot; 3902 int index; 3903 bool hotplugged = spapr_drc_hotplugged(dev); 3904 int i; 3905 3906 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index); 3907 g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */ 3908 3909 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, 3910 spapr_vcpu_id(spapr, cc->core_id)); 3911 3912 g_assert(drc || !mc->has_hotpluggable_cpus); 3913 3914 if (drc) { 3915 /* 3916 * spapr_core_pre_plug() already buys us this is a brand new 3917 * core being plugged into a free slot. Nothing should already 3918 * be attached to the corresponding DRC. 3919 */ 3920 spapr_drc_attach(drc, dev); 3921 3922 if (hotplugged) { 3923 /* 3924 * Send hotplug notification interrupt to the guest only 3925 * in case of hotplugged CPUs. 3926 */ 3927 spapr_hotplug_req_add_by_index(drc); 3928 } else { 3929 spapr_drc_reset(drc); 3930 } 3931 } 3932 3933 core_slot->cpu = OBJECT(dev); 3934 3935 /* 3936 * Set compatibility mode to match the boot CPU, which was either set 3937 * by the machine reset code or by CAS. This really shouldn't fail at 3938 * this point. 3939 */ 3940 if (hotplugged) { 3941 for (i = 0; i < cc->nr_threads; i++) { 3942 ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr, 3943 &error_abort); 3944 } 3945 } 3946 3947 if (smc->pre_2_10_has_unused_icps) { 3948 for (i = 0; i < cc->nr_threads; i++) { 3949 cs = CPU(core->threads[i]); 3950 pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index); 3951 } 3952 } 3953 } 3954 3955 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, 3956 Error **errp) 3957 { 3958 MachineState *machine = MACHINE(OBJECT(hotplug_dev)); 3959 MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev); 3960 CPUCore *cc = CPU_CORE(dev); 3961 const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type); 3962 const char *type = object_get_typename(OBJECT(dev)); 3963 CPUArchId *core_slot; 3964 int index; 3965 unsigned int smp_threads = machine->smp.threads; 3966 3967 if (dev->hotplugged && !mc->has_hotpluggable_cpus) { 3968 error_setg(errp, "CPU hotplug not supported for this machine"); 3969 return; 3970 } 3971 3972 if (strcmp(base_core_type, type)) { 3973 error_setg(errp, "CPU core type should be %s", base_core_type); 3974 return; 3975 } 3976 3977 if (cc->core_id % smp_threads) { 3978 error_setg(errp, "invalid core id %d", cc->core_id); 3979 return; 3980 } 3981 3982 /* 3983 * In general we should have homogeneous threads-per-core, but old 3984 * (pre hotplug support) machine types allow the last core to have 3985 * reduced threads as a compatibility hack for when we allowed 3986 * total vcpus not a multiple of threads-per-core. 3987 */ 3988 if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) { 3989 error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads, 3990 smp_threads); 3991 return; 3992 } 3993 3994 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index); 3995 if (!core_slot) { 3996 error_setg(errp, "core id %d out of range", cc->core_id); 3997 return; 3998 } 3999 4000 if (core_slot->cpu) { 4001 error_setg(errp, "core %d already populated", cc->core_id); 4002 return; 4003 } 4004 4005 numa_cpu_pre_plug(core_slot, dev, errp); 4006 } 4007 4008 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr, 4009 void *fdt, int *fdt_start_offset, Error **errp) 4010 { 4011 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev); 4012 int intc_phandle; 4013 4014 intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp); 4015 if (intc_phandle <= 0) { 4016 return -1; 4017 } 4018 4019 if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) { 4020 error_setg(errp, "unable to create FDT node for PHB %d", sphb->index); 4021 return -1; 4022 } 4023 4024 /* generally SLOF creates these, for hotplug it's up to QEMU */ 4025 _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci")); 4026 4027 return 0; 4028 } 4029 4030 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, 4031 Error **errp) 4032 { 4033 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev)); 4034 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev); 4035 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 4036 const unsigned windows_supported = spapr_phb_windows_supported(sphb); 4037 SpaprDrc *drc; 4038 4039 if (dev->hotplugged && !smc->dr_phb_enabled) { 4040 error_setg(errp, "PHB hotplug not supported for this machine"); 4041 return false; 4042 } 4043 4044 if (sphb->index == (uint32_t)-1) { 4045 error_setg(errp, "\"index\" for PAPR PHB is mandatory"); 4046 return false; 4047 } 4048 4049 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index); 4050 if (drc && drc->dev) { 4051 error_setg(errp, "PHB %d already attached", sphb->index); 4052 return false; 4053 } 4054 4055 /* 4056 * This will check that sphb->index doesn't exceed the maximum number of 4057 * PHBs for the current machine type. 4058 */ 4059 return 4060 smc->phb_placement(spapr, sphb->index, 4061 &sphb->buid, &sphb->io_win_addr, 4062 &sphb->mem_win_addr, &sphb->mem64_win_addr, 4063 windows_supported, sphb->dma_liobn, 4064 &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr, 4065 errp); 4066 } 4067 4068 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev) 4069 { 4070 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev)); 4071 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 4072 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev); 4073 SpaprDrc *drc; 4074 bool hotplugged = spapr_drc_hotplugged(dev); 4075 4076 if (!smc->dr_phb_enabled) { 4077 return; 4078 } 4079 4080 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index); 4081 /* hotplug hooks should check it's enabled before getting this far */ 4082 assert(drc); 4083 4084 /* spapr_phb_pre_plug() already checked the DRC is attachable */ 4085 spapr_drc_attach(drc, dev); 4086 4087 if (hotplugged) { 4088 spapr_hotplug_req_add_by_index(drc); 4089 } else { 4090 spapr_drc_reset(drc); 4091 } 4092 } 4093 4094 void spapr_phb_release(DeviceState *dev) 4095 { 4096 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev); 4097 4098 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort); 4099 object_unparent(OBJECT(dev)); 4100 } 4101 4102 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev) 4103 { 4104 qdev_unrealize(dev); 4105 } 4106 4107 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev, 4108 DeviceState *dev, Error **errp) 4109 { 4110 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev); 4111 SpaprDrc *drc; 4112 4113 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index); 4114 assert(drc); 4115 4116 if (!spapr_drc_unplug_requested(drc)) { 4117 spapr_drc_unplug_request(drc); 4118 spapr_hotplug_req_remove_by_index(drc); 4119 } else { 4120 error_setg(errp, 4121 "PCI Host Bridge unplug already in progress for device %s", 4122 dev->id); 4123 } 4124 } 4125 4126 static 4127 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, 4128 Error **errp) 4129 { 4130 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev)); 4131 4132 if (spapr->tpm_proxy != NULL) { 4133 error_setg(errp, "Only one TPM proxy can be specified for this machine"); 4134 return false; 4135 } 4136 4137 return true; 4138 } 4139 4140 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev) 4141 { 4142 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev)); 4143 SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev); 4144 4145 /* Already checked in spapr_tpm_proxy_pre_plug() */ 4146 g_assert(spapr->tpm_proxy == NULL); 4147 4148 spapr->tpm_proxy = tpm_proxy; 4149 } 4150 4151 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev) 4152 { 4153 SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev)); 4154 4155 qdev_unrealize(dev); 4156 object_unparent(OBJECT(dev)); 4157 spapr->tpm_proxy = NULL; 4158 } 4159 4160 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev, 4161 DeviceState *dev, Error **errp) 4162 { 4163 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) { 4164 spapr_memory_plug(hotplug_dev, dev); 4165 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) { 4166 spapr_core_plug(hotplug_dev, dev); 4167 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) { 4168 spapr_phb_plug(hotplug_dev, dev); 4169 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) { 4170 spapr_tpm_proxy_plug(hotplug_dev, dev); 4171 } 4172 } 4173 4174 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev, 4175 DeviceState *dev, Error **errp) 4176 { 4177 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) { 4178 spapr_memory_unplug(hotplug_dev, dev); 4179 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) { 4180 spapr_core_unplug(hotplug_dev, dev); 4181 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) { 4182 spapr_phb_unplug(hotplug_dev, dev); 4183 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) { 4184 spapr_tpm_proxy_unplug(hotplug_dev, dev); 4185 } 4186 } 4187 4188 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr) 4189 { 4190 return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) || 4191 /* 4192 * CAS will process all pending unplug requests. 4193 * 4194 * HACK: a guest could theoretically have cleared all bits in OV5, 4195 * but none of the guests we care for do. 4196 */ 4197 spapr_ovec_empty(spapr->ov5_cas); 4198 } 4199 4200 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev, 4201 DeviceState *dev, Error **errp) 4202 { 4203 SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev)); 4204 MachineClass *mc = MACHINE_GET_CLASS(sms); 4205 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4206 4207 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) { 4208 if (spapr_memory_hot_unplug_supported(sms)) { 4209 spapr_memory_unplug_request(hotplug_dev, dev, errp); 4210 } else { 4211 error_setg(errp, "Memory hot unplug not supported for this guest"); 4212 } 4213 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) { 4214 if (!mc->has_hotpluggable_cpus) { 4215 error_setg(errp, "CPU hot unplug not supported on this machine"); 4216 return; 4217 } 4218 spapr_core_unplug_request(hotplug_dev, dev, errp); 4219 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) { 4220 if (!smc->dr_phb_enabled) { 4221 error_setg(errp, "PHB hot unplug not supported on this machine"); 4222 return; 4223 } 4224 spapr_phb_unplug_request(hotplug_dev, dev, errp); 4225 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) { 4226 spapr_tpm_proxy_unplug(hotplug_dev, dev); 4227 } 4228 } 4229 4230 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev, 4231 DeviceState *dev, Error **errp) 4232 { 4233 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) { 4234 spapr_memory_pre_plug(hotplug_dev, dev, errp); 4235 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) { 4236 spapr_core_pre_plug(hotplug_dev, dev, errp); 4237 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) { 4238 spapr_phb_pre_plug(hotplug_dev, dev, errp); 4239 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) { 4240 spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp); 4241 } 4242 } 4243 4244 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine, 4245 DeviceState *dev) 4246 { 4247 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) || 4248 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) || 4249 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) || 4250 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) { 4251 return HOTPLUG_HANDLER(machine); 4252 } 4253 if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) { 4254 PCIDevice *pcidev = PCI_DEVICE(dev); 4255 PCIBus *root = pci_device_root_bus(pcidev); 4256 SpaprPhbState *phb = 4257 (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent), 4258 TYPE_SPAPR_PCI_HOST_BRIDGE); 4259 4260 if (phb) { 4261 return HOTPLUG_HANDLER(phb); 4262 } 4263 } 4264 return NULL; 4265 } 4266 4267 static CpuInstanceProperties 4268 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index) 4269 { 4270 CPUArchId *core_slot; 4271 MachineClass *mc = MACHINE_GET_CLASS(machine); 4272 4273 /* make sure possible_cpu are intialized */ 4274 mc->possible_cpu_arch_ids(machine); 4275 /* get CPU core slot containing thread that matches cpu_index */ 4276 core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL); 4277 assert(core_slot); 4278 return core_slot->props; 4279 } 4280 4281 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx) 4282 { 4283 return idx / ms->smp.cores % ms->numa_state->num_nodes; 4284 } 4285 4286 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine) 4287 { 4288 int i; 4289 unsigned int smp_threads = machine->smp.threads; 4290 unsigned int smp_cpus = machine->smp.cpus; 4291 const char *core_type; 4292 int spapr_max_cores = machine->smp.max_cpus / smp_threads; 4293 MachineClass *mc = MACHINE_GET_CLASS(machine); 4294 4295 if (!mc->has_hotpluggable_cpus) { 4296 spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads; 4297 } 4298 if (machine->possible_cpus) { 4299 assert(machine->possible_cpus->len == spapr_max_cores); 4300 return machine->possible_cpus; 4301 } 4302 4303 core_type = spapr_get_cpu_core_type(machine->cpu_type); 4304 if (!core_type) { 4305 error_report("Unable to find sPAPR CPU Core definition"); 4306 exit(1); 4307 } 4308 4309 machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) + 4310 sizeof(CPUArchId) * spapr_max_cores); 4311 machine->possible_cpus->len = spapr_max_cores; 4312 for (i = 0; i < machine->possible_cpus->len; i++) { 4313 int core_id = i * smp_threads; 4314 4315 machine->possible_cpus->cpus[i].type = core_type; 4316 machine->possible_cpus->cpus[i].vcpus_count = smp_threads; 4317 machine->possible_cpus->cpus[i].arch_id = core_id; 4318 machine->possible_cpus->cpus[i].props.has_core_id = true; 4319 machine->possible_cpus->cpus[i].props.core_id = core_id; 4320 } 4321 return machine->possible_cpus; 4322 } 4323 4324 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index, 4325 uint64_t *buid, hwaddr *pio, 4326 hwaddr *mmio32, hwaddr *mmio64, 4327 unsigned n_dma, uint32_t *liobns, 4328 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp) 4329 { 4330 /* 4331 * New-style PHB window placement. 4332 * 4333 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window 4334 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO 4335 * windows. 4336 * 4337 * Some guest kernels can't work with MMIO windows above 1<<46 4338 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB 4339 * 4340 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each 4341 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the 4342 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the 4343 * 1TiB 64-bit MMIO windows for each PHB. 4344 */ 4345 const uint64_t base_buid = 0x800000020000000ULL; 4346 int i; 4347 4348 /* Sanity check natural alignments */ 4349 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0); 4350 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0); 4351 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0); 4352 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0); 4353 /* Sanity check bounds */ 4354 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) > 4355 SPAPR_PCI_MEM32_WIN_SIZE); 4356 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) > 4357 SPAPR_PCI_MEM64_WIN_SIZE); 4358 4359 if (index >= SPAPR_MAX_PHBS) { 4360 error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)", 4361 SPAPR_MAX_PHBS - 1); 4362 return false; 4363 } 4364 4365 *buid = base_buid + index; 4366 for (i = 0; i < n_dma; ++i) { 4367 liobns[i] = SPAPR_PCI_LIOBN(index, i); 4368 } 4369 4370 *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE; 4371 *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE; 4372 *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE; 4373 4374 *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE; 4375 *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE; 4376 return true; 4377 } 4378 4379 static ICSState *spapr_ics_get(XICSFabric *dev, int irq) 4380 { 4381 SpaprMachineState *spapr = SPAPR_MACHINE(dev); 4382 4383 return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL; 4384 } 4385 4386 static void spapr_ics_resend(XICSFabric *dev) 4387 { 4388 SpaprMachineState *spapr = SPAPR_MACHINE(dev); 4389 4390 ics_resend(spapr->ics); 4391 } 4392 4393 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id) 4394 { 4395 PowerPCCPU *cpu = spapr_find_cpu(vcpu_id); 4396 4397 return cpu ? spapr_cpu_state(cpu)->icp : NULL; 4398 } 4399 4400 static void spapr_pic_print_info(InterruptStatsProvider *obj, 4401 Monitor *mon) 4402 { 4403 SpaprMachineState *spapr = SPAPR_MACHINE(obj); 4404 4405 spapr_irq_print_info(spapr, mon); 4406 monitor_printf(mon, "irqchip: %s\n", 4407 kvm_irqchip_in_kernel() ? "in-kernel" : "emulated"); 4408 } 4409 4410 /* 4411 * This is a XIVE only operation 4412 */ 4413 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format, 4414 uint8_t nvt_blk, uint32_t nvt_idx, 4415 bool cam_ignore, uint8_t priority, 4416 uint32_t logic_serv, XiveTCTXMatch *match) 4417 { 4418 SpaprMachineState *spapr = SPAPR_MACHINE(xfb); 4419 XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc); 4420 XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr); 4421 int count; 4422 4423 count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore, 4424 priority, logic_serv, match); 4425 if (count < 0) { 4426 return count; 4427 } 4428 4429 /* 4430 * When we implement the save and restore of the thread interrupt 4431 * contexts in the enter/exit CPU handlers of the machine and the 4432 * escalations in QEMU, we should be able to handle non dispatched 4433 * vCPUs. 4434 * 4435 * Until this is done, the sPAPR machine should find at least one 4436 * matching context always. 4437 */ 4438 if (count == 0) { 4439 qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n", 4440 nvt_blk, nvt_idx); 4441 } 4442 4443 return count; 4444 } 4445 4446 int spapr_get_vcpu_id(PowerPCCPU *cpu) 4447 { 4448 return cpu->vcpu_id; 4449 } 4450 4451 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp) 4452 { 4453 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 4454 MachineState *ms = MACHINE(spapr); 4455 int vcpu_id; 4456 4457 vcpu_id = spapr_vcpu_id(spapr, cpu_index); 4458 4459 if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) { 4460 error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id); 4461 error_append_hint(errp, "Adjust the number of cpus to %d " 4462 "or try to raise the number of threads per core\n", 4463 vcpu_id * ms->smp.threads / spapr->vsmt); 4464 return false; 4465 } 4466 4467 cpu->vcpu_id = vcpu_id; 4468 return true; 4469 } 4470 4471 PowerPCCPU *spapr_find_cpu(int vcpu_id) 4472 { 4473 CPUState *cs; 4474 4475 CPU_FOREACH(cs) { 4476 PowerPCCPU *cpu = POWERPC_CPU(cs); 4477 4478 if (spapr_get_vcpu_id(cpu) == vcpu_id) { 4479 return cpu; 4480 } 4481 } 4482 4483 return NULL; 4484 } 4485 4486 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu) 4487 { 4488 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu); 4489 4490 /* These are only called by TCG, KVM maintains dispatch state */ 4491 4492 spapr_cpu->prod = false; 4493 if (spapr_cpu->vpa_addr) { 4494 CPUState *cs = CPU(cpu); 4495 uint32_t dispatch; 4496 4497 dispatch = ldl_be_phys(cs->as, 4498 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER); 4499 dispatch++; 4500 if ((dispatch & 1) != 0) { 4501 qemu_log_mask(LOG_GUEST_ERROR, 4502 "VPA: incorrect dispatch counter value for " 4503 "dispatched partition %u, correcting.\n", dispatch); 4504 dispatch++; 4505 } 4506 stl_be_phys(cs->as, 4507 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch); 4508 } 4509 } 4510 4511 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu) 4512 { 4513 SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu); 4514 4515 if (spapr_cpu->vpa_addr) { 4516 CPUState *cs = CPU(cpu); 4517 uint32_t dispatch; 4518 4519 dispatch = ldl_be_phys(cs->as, 4520 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER); 4521 dispatch++; 4522 if ((dispatch & 1) != 1) { 4523 qemu_log_mask(LOG_GUEST_ERROR, 4524 "VPA: incorrect dispatch counter value for " 4525 "preempted partition %u, correcting.\n", dispatch); 4526 dispatch++; 4527 } 4528 stl_be_phys(cs->as, 4529 spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch); 4530 } 4531 } 4532 4533 static void spapr_machine_class_init(ObjectClass *oc, void *data) 4534 { 4535 MachineClass *mc = MACHINE_CLASS(oc); 4536 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc); 4537 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc); 4538 NMIClass *nc = NMI_CLASS(oc); 4539 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc); 4540 PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc); 4541 XICSFabricClass *xic = XICS_FABRIC_CLASS(oc); 4542 InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc); 4543 XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc); 4544 VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc); 4545 4546 mc->desc = "pSeries Logical Partition (PAPR compliant)"; 4547 mc->ignore_boot_device_suffixes = true; 4548 4549 /* 4550 * We set up the default / latest behaviour here. The class_init 4551 * functions for the specific versioned machine types can override 4552 * these details for backwards compatibility 4553 */ 4554 mc->init = spapr_machine_init; 4555 mc->reset = spapr_machine_reset; 4556 mc->block_default_type = IF_SCSI; 4557 4558 /* 4559 * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values 4560 * should be limited by the host capability instead of hardcoded. 4561 * max_cpus for KVM guests will be checked in kvm_init(), and TCG 4562 * guests are welcome to have as many CPUs as the host are capable 4563 * of emulate. 4564 */ 4565 mc->max_cpus = INT32_MAX; 4566 4567 mc->no_parallel = 1; 4568 mc->default_boot_order = ""; 4569 mc->default_ram_size = 512 * MiB; 4570 mc->default_ram_id = "ppc_spapr.ram"; 4571 mc->default_display = "std"; 4572 mc->kvm_type = spapr_kvm_type; 4573 machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE); 4574 mc->pci_allow_0_address = true; 4575 assert(!mc->get_hotplug_handler); 4576 mc->get_hotplug_handler = spapr_get_hotplug_handler; 4577 hc->pre_plug = spapr_machine_device_pre_plug; 4578 hc->plug = spapr_machine_device_plug; 4579 mc->cpu_index_to_instance_props = spapr_cpu_index_to_props; 4580 mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id; 4581 mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids; 4582 hc->unplug_request = spapr_machine_device_unplug_request; 4583 hc->unplug = spapr_machine_device_unplug; 4584 4585 smc->dr_lmb_enabled = true; 4586 smc->update_dt_enabled = true; 4587 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0"); 4588 mc->has_hotpluggable_cpus = true; 4589 mc->nvdimm_supported = true; 4590 smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED; 4591 fwc->get_dev_path = spapr_get_fw_dev_path; 4592 nc->nmi_monitor_handler = spapr_nmi; 4593 smc->phb_placement = spapr_phb_placement; 4594 vhc->hypercall = emulate_spapr_hypercall; 4595 vhc->hpt_mask = spapr_hpt_mask; 4596 vhc->map_hptes = spapr_map_hptes; 4597 vhc->unmap_hptes = spapr_unmap_hptes; 4598 vhc->hpte_set_c = spapr_hpte_set_c; 4599 vhc->hpte_set_r = spapr_hpte_set_r; 4600 vhc->get_pate = spapr_get_pate; 4601 vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr; 4602 vhc->cpu_exec_enter = spapr_cpu_exec_enter; 4603 vhc->cpu_exec_exit = spapr_cpu_exec_exit; 4604 xic->ics_get = spapr_ics_get; 4605 xic->ics_resend = spapr_ics_resend; 4606 xic->icp_get = spapr_icp_get; 4607 ispc->print_info = spapr_pic_print_info; 4608 /* Force NUMA node memory size to be a multiple of 4609 * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity 4610 * in which LMBs are represented and hot-added 4611 */ 4612 mc->numa_mem_align_shift = 28; 4613 mc->auto_enable_numa = true; 4614 4615 smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF; 4616 smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON; 4617 smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON; 4618 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND; 4619 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND; 4620 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND; 4621 smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */ 4622 smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF; 4623 smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON; 4624 smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON; 4625 smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON; 4626 smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF; 4627 spapr_caps_add_properties(smc); 4628 smc->irq = &spapr_irq_dual; 4629 smc->dr_phb_enabled = true; 4630 smc->linux_pci_probe = true; 4631 smc->smp_threads_vsmt = true; 4632 smc->nr_xirqs = SPAPR_NR_XIRQS; 4633 xfc->match_nvt = spapr_match_nvt; 4634 vmc->client_architecture_support = spapr_vof_client_architecture_support; 4635 vmc->quiesce = spapr_vof_quiesce; 4636 vmc->setprop = spapr_vof_setprop; 4637 } 4638 4639 static const TypeInfo spapr_machine_info = { 4640 .name = TYPE_SPAPR_MACHINE, 4641 .parent = TYPE_MACHINE, 4642 .abstract = true, 4643 .instance_size = sizeof(SpaprMachineState), 4644 .instance_init = spapr_instance_init, 4645 .instance_finalize = spapr_machine_finalizefn, 4646 .class_size = sizeof(SpaprMachineClass), 4647 .class_init = spapr_machine_class_init, 4648 .interfaces = (InterfaceInfo[]) { 4649 { TYPE_FW_PATH_PROVIDER }, 4650 { TYPE_NMI }, 4651 { TYPE_HOTPLUG_HANDLER }, 4652 { TYPE_PPC_VIRTUAL_HYPERVISOR }, 4653 { TYPE_XICS_FABRIC }, 4654 { TYPE_INTERRUPT_STATS_PROVIDER }, 4655 { TYPE_XIVE_FABRIC }, 4656 { TYPE_VOF_MACHINE_IF }, 4657 { } 4658 }, 4659 }; 4660 4661 static void spapr_machine_latest_class_options(MachineClass *mc) 4662 { 4663 mc->alias = "pseries"; 4664 mc->is_default = true; 4665 } 4666 4667 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \ 4668 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \ 4669 void *data) \ 4670 { \ 4671 MachineClass *mc = MACHINE_CLASS(oc); \ 4672 spapr_machine_##suffix##_class_options(mc); \ 4673 if (latest) { \ 4674 spapr_machine_latest_class_options(mc); \ 4675 } \ 4676 } \ 4677 static const TypeInfo spapr_machine_##suffix##_info = { \ 4678 .name = MACHINE_TYPE_NAME("pseries-" verstr), \ 4679 .parent = TYPE_SPAPR_MACHINE, \ 4680 .class_init = spapr_machine_##suffix##_class_init, \ 4681 }; \ 4682 static void spapr_machine_register_##suffix(void) \ 4683 { \ 4684 type_register(&spapr_machine_##suffix##_info); \ 4685 } \ 4686 type_init(spapr_machine_register_##suffix) 4687 4688 /* 4689 * pseries-6.2 4690 */ 4691 static void spapr_machine_6_2_class_options(MachineClass *mc) 4692 { 4693 /* Defaults for the latest behaviour inherited from the base class */ 4694 } 4695 4696 DEFINE_SPAPR_MACHINE(6_2, "6.2", true); 4697 4698 /* 4699 * pseries-6.1 4700 */ 4701 static void spapr_machine_6_1_class_options(MachineClass *mc) 4702 { 4703 spapr_machine_6_2_class_options(mc); 4704 compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len); 4705 } 4706 4707 DEFINE_SPAPR_MACHINE(6_1, "6.1", false); 4708 4709 /* 4710 * pseries-6.0 4711 */ 4712 static void spapr_machine_6_0_class_options(MachineClass *mc) 4713 { 4714 spapr_machine_6_1_class_options(mc); 4715 compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len); 4716 } 4717 4718 DEFINE_SPAPR_MACHINE(6_0, "6.0", false); 4719 4720 /* 4721 * pseries-5.2 4722 */ 4723 static void spapr_machine_5_2_class_options(MachineClass *mc) 4724 { 4725 spapr_machine_6_0_class_options(mc); 4726 compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len); 4727 } 4728 4729 DEFINE_SPAPR_MACHINE(5_2, "5.2", false); 4730 4731 /* 4732 * pseries-5.1 4733 */ 4734 static void spapr_machine_5_1_class_options(MachineClass *mc) 4735 { 4736 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4737 4738 spapr_machine_5_2_class_options(mc); 4739 compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len); 4740 smc->pre_5_2_numa_associativity = true; 4741 } 4742 4743 DEFINE_SPAPR_MACHINE(5_1, "5.1", false); 4744 4745 /* 4746 * pseries-5.0 4747 */ 4748 static void spapr_machine_5_0_class_options(MachineClass *mc) 4749 { 4750 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4751 static GlobalProperty compat[] = { 4752 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" }, 4753 }; 4754 4755 spapr_machine_5_1_class_options(mc); 4756 compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len); 4757 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 4758 mc->numa_mem_supported = true; 4759 smc->pre_5_1_assoc_refpoints = true; 4760 } 4761 4762 DEFINE_SPAPR_MACHINE(5_0, "5.0", false); 4763 4764 /* 4765 * pseries-4.2 4766 */ 4767 static void spapr_machine_4_2_class_options(MachineClass *mc) 4768 { 4769 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4770 4771 spapr_machine_5_0_class_options(mc); 4772 compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len); 4773 smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF; 4774 smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF; 4775 smc->rma_limit = 16 * GiB; 4776 mc->nvdimm_supported = false; 4777 } 4778 4779 DEFINE_SPAPR_MACHINE(4_2, "4.2", false); 4780 4781 /* 4782 * pseries-4.1 4783 */ 4784 static void spapr_machine_4_1_class_options(MachineClass *mc) 4785 { 4786 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4787 static GlobalProperty compat[] = { 4788 /* Only allow 4kiB and 64kiB IOMMU pagesizes */ 4789 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" }, 4790 }; 4791 4792 spapr_machine_4_2_class_options(mc); 4793 smc->linux_pci_probe = false; 4794 smc->smp_threads_vsmt = false; 4795 compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len); 4796 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 4797 } 4798 4799 DEFINE_SPAPR_MACHINE(4_1, "4.1", false); 4800 4801 /* 4802 * pseries-4.0 4803 */ 4804 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index, 4805 uint64_t *buid, hwaddr *pio, 4806 hwaddr *mmio32, hwaddr *mmio64, 4807 unsigned n_dma, uint32_t *liobns, 4808 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp) 4809 { 4810 if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, 4811 liobns, nv2gpa, nv2atsd, errp)) { 4812 return false; 4813 } 4814 4815 *nv2gpa = 0; 4816 *nv2atsd = 0; 4817 return true; 4818 } 4819 static void spapr_machine_4_0_class_options(MachineClass *mc) 4820 { 4821 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4822 4823 spapr_machine_4_1_class_options(mc); 4824 compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len); 4825 smc->phb_placement = phb_placement_4_0; 4826 smc->irq = &spapr_irq_xics; 4827 smc->pre_4_1_migration = true; 4828 } 4829 4830 DEFINE_SPAPR_MACHINE(4_0, "4.0", false); 4831 4832 /* 4833 * pseries-3.1 4834 */ 4835 static void spapr_machine_3_1_class_options(MachineClass *mc) 4836 { 4837 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4838 4839 spapr_machine_4_0_class_options(mc); 4840 compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len); 4841 4842 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0"); 4843 smc->update_dt_enabled = false; 4844 smc->dr_phb_enabled = false; 4845 smc->broken_host_serial_model = true; 4846 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN; 4847 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN; 4848 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN; 4849 smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF; 4850 } 4851 4852 DEFINE_SPAPR_MACHINE(3_1, "3.1", false); 4853 4854 /* 4855 * pseries-3.0 4856 */ 4857 4858 static void spapr_machine_3_0_class_options(MachineClass *mc) 4859 { 4860 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4861 4862 spapr_machine_3_1_class_options(mc); 4863 compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len); 4864 4865 smc->legacy_irq_allocation = true; 4866 smc->nr_xirqs = 0x400; 4867 smc->irq = &spapr_irq_xics_legacy; 4868 } 4869 4870 DEFINE_SPAPR_MACHINE(3_0, "3.0", false); 4871 4872 /* 4873 * pseries-2.12 4874 */ 4875 static void spapr_machine_2_12_class_options(MachineClass *mc) 4876 { 4877 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4878 static GlobalProperty compat[] = { 4879 { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" }, 4880 { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" }, 4881 }; 4882 4883 spapr_machine_3_0_class_options(mc); 4884 compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len); 4885 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 4886 4887 /* We depend on kvm_enabled() to choose a default value for the 4888 * hpt-max-page-size capability. Of course we can't do it here 4889 * because this is too early and the HW accelerator isn't initialzed 4890 * yet. Postpone this to machine init (see default_caps_with_cpu()). 4891 */ 4892 smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0; 4893 } 4894 4895 DEFINE_SPAPR_MACHINE(2_12, "2.12", false); 4896 4897 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc) 4898 { 4899 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4900 4901 spapr_machine_2_12_class_options(mc); 4902 smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND; 4903 smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND; 4904 smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD; 4905 } 4906 4907 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false); 4908 4909 /* 4910 * pseries-2.11 4911 */ 4912 4913 static void spapr_machine_2_11_class_options(MachineClass *mc) 4914 { 4915 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4916 4917 spapr_machine_2_12_class_options(mc); 4918 smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON; 4919 compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len); 4920 } 4921 4922 DEFINE_SPAPR_MACHINE(2_11, "2.11", false); 4923 4924 /* 4925 * pseries-2.10 4926 */ 4927 4928 static void spapr_machine_2_10_class_options(MachineClass *mc) 4929 { 4930 spapr_machine_2_11_class_options(mc); 4931 compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len); 4932 } 4933 4934 DEFINE_SPAPR_MACHINE(2_10, "2.10", false); 4935 4936 /* 4937 * pseries-2.9 4938 */ 4939 4940 static void spapr_machine_2_9_class_options(MachineClass *mc) 4941 { 4942 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 4943 static GlobalProperty compat[] = { 4944 { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" }, 4945 }; 4946 4947 spapr_machine_2_10_class_options(mc); 4948 compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len); 4949 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 4950 smc->pre_2_10_has_unused_icps = true; 4951 smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED; 4952 } 4953 4954 DEFINE_SPAPR_MACHINE(2_9, "2.9", false); 4955 4956 /* 4957 * pseries-2.8 4958 */ 4959 4960 static void spapr_machine_2_8_class_options(MachineClass *mc) 4961 { 4962 static GlobalProperty compat[] = { 4963 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" }, 4964 }; 4965 4966 spapr_machine_2_9_class_options(mc); 4967 compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len); 4968 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 4969 mc->numa_mem_align_shift = 23; 4970 } 4971 4972 DEFINE_SPAPR_MACHINE(2_8, "2.8", false); 4973 4974 /* 4975 * pseries-2.7 4976 */ 4977 4978 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index, 4979 uint64_t *buid, hwaddr *pio, 4980 hwaddr *mmio32, hwaddr *mmio64, 4981 unsigned n_dma, uint32_t *liobns, 4982 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp) 4983 { 4984 /* Legacy PHB placement for pseries-2.7 and earlier machine types */ 4985 const uint64_t base_buid = 0x800000020000000ULL; 4986 const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */ 4987 const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */ 4988 const hwaddr pio_offset = 0x80000000; /* 2 GiB */ 4989 const uint32_t max_index = 255; 4990 const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */ 4991 4992 uint64_t ram_top = MACHINE(spapr)->ram_size; 4993 hwaddr phb0_base, phb_base; 4994 int i; 4995 4996 /* Do we have device memory? */ 4997 if (MACHINE(spapr)->maxram_size > ram_top) { 4998 /* Can't just use maxram_size, because there may be an 4999 * alignment gap between normal and device memory regions 5000 */ 5001 ram_top = MACHINE(spapr)->device_memory->base + 5002 memory_region_size(&MACHINE(spapr)->device_memory->mr); 5003 } 5004 5005 phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment); 5006 5007 if (index > max_index) { 5008 error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)", 5009 max_index); 5010 return false; 5011 } 5012 5013 *buid = base_buid + index; 5014 for (i = 0; i < n_dma; ++i) { 5015 liobns[i] = SPAPR_PCI_LIOBN(index, i); 5016 } 5017 5018 phb_base = phb0_base + index * phb_spacing; 5019 *pio = phb_base + pio_offset; 5020 *mmio32 = phb_base + mmio_offset; 5021 /* 5022 * We don't set the 64-bit MMIO window, relying on the PHB's 5023 * fallback behaviour of automatically splitting a large "32-bit" 5024 * window into contiguous 32-bit and 64-bit windows 5025 */ 5026 5027 *nv2gpa = 0; 5028 *nv2atsd = 0; 5029 return true; 5030 } 5031 5032 static void spapr_machine_2_7_class_options(MachineClass *mc) 5033 { 5034 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 5035 static GlobalProperty compat[] = { 5036 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", }, 5037 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", }, 5038 { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", }, 5039 { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", }, 5040 }; 5041 5042 spapr_machine_2_8_class_options(mc); 5043 mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3"); 5044 mc->default_machine_opts = "modern-hotplug-events=off"; 5045 compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len); 5046 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 5047 smc->phb_placement = phb_placement_2_7; 5048 } 5049 5050 DEFINE_SPAPR_MACHINE(2_7, "2.7", false); 5051 5052 /* 5053 * pseries-2.6 5054 */ 5055 5056 static void spapr_machine_2_6_class_options(MachineClass *mc) 5057 { 5058 static GlobalProperty compat[] = { 5059 { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" }, 5060 }; 5061 5062 spapr_machine_2_7_class_options(mc); 5063 mc->has_hotpluggable_cpus = false; 5064 compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len); 5065 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 5066 } 5067 5068 DEFINE_SPAPR_MACHINE(2_6, "2.6", false); 5069 5070 /* 5071 * pseries-2.5 5072 */ 5073 5074 static void spapr_machine_2_5_class_options(MachineClass *mc) 5075 { 5076 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 5077 static GlobalProperty compat[] = { 5078 { "spapr-vlan", "use-rx-buffer-pools", "off" }, 5079 }; 5080 5081 spapr_machine_2_6_class_options(mc); 5082 smc->use_ohci_by_default = true; 5083 compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len); 5084 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 5085 } 5086 5087 DEFINE_SPAPR_MACHINE(2_5, "2.5", false); 5088 5089 /* 5090 * pseries-2.4 5091 */ 5092 5093 static void spapr_machine_2_4_class_options(MachineClass *mc) 5094 { 5095 SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc); 5096 5097 spapr_machine_2_5_class_options(mc); 5098 smc->dr_lmb_enabled = false; 5099 compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len); 5100 } 5101 5102 DEFINE_SPAPR_MACHINE(2_4, "2.4", false); 5103 5104 /* 5105 * pseries-2.3 5106 */ 5107 5108 static void spapr_machine_2_3_class_options(MachineClass *mc) 5109 { 5110 static GlobalProperty compat[] = { 5111 { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" }, 5112 }; 5113 spapr_machine_2_4_class_options(mc); 5114 compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len); 5115 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 5116 } 5117 DEFINE_SPAPR_MACHINE(2_3, "2.3", false); 5118 5119 /* 5120 * pseries-2.2 5121 */ 5122 5123 static void spapr_machine_2_2_class_options(MachineClass *mc) 5124 { 5125 static GlobalProperty compat[] = { 5126 { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" }, 5127 }; 5128 5129 spapr_machine_2_3_class_options(mc); 5130 compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len); 5131 compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat)); 5132 mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on"; 5133 } 5134 DEFINE_SPAPR_MACHINE(2_2, "2.2", false); 5135 5136 /* 5137 * pseries-2.1 5138 */ 5139 5140 static void spapr_machine_2_1_class_options(MachineClass *mc) 5141 { 5142 spapr_machine_2_2_class_options(mc); 5143 compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len); 5144 } 5145 DEFINE_SPAPR_MACHINE(2_1, "2.1", false); 5146 5147 static void spapr_machine_register_types(void) 5148 { 5149 type_register_static(&spapr_machine_info); 5150 } 5151 5152 type_init(spapr_machine_register_types) 5153