1 /* 2 * Copyright (c) 2003-2004 Fabrice Bellard 3 * Copyright (c) 2019 Red Hat, Inc. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a copy 6 * of this software and associated documentation files (the "Software"), to deal 7 * in the Software without restriction, including without limitation the rights 8 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 * copies of the Software, and to permit persons to whom the Software is 10 * furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice shall be included in 13 * all copies or substantial portions of the Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 21 * THE SOFTWARE. 22 */ 23 #include "qemu/osdep.h" 24 #include "qemu/error-report.h" 25 #include "qemu/option.h" 26 #include "qemu/cutils.h" 27 #include "qemu/units.h" 28 #include "qemu/datadir.h" 29 #include "qapi/error.h" 30 #include "qapi/qapi-visit-common.h" 31 #include "qapi/clone-visitor.h" 32 #include "qapi/qapi-visit-machine.h" 33 #include "qapi/visitor.h" 34 #include "sysemu/qtest.h" 35 #include "sysemu/whpx.h" 36 #include "sysemu/numa.h" 37 #include "sysemu/replay.h" 38 #include "sysemu/sysemu.h" 39 #include "sysemu/cpu-timers.h" 40 #include "sysemu/xen.h" 41 #include "trace.h" 42 43 #include "hw/i386/x86.h" 44 #include "target/i386/cpu.h" 45 #include "hw/i386/topology.h" 46 #include "hw/i386/fw_cfg.h" 47 #include "hw/intc/i8259.h" 48 #include "hw/rtc/mc146818rtc.h" 49 #include "target/i386/sev.h" 50 51 #include "hw/acpi/cpu_hotplug.h" 52 #include "hw/irq.h" 53 #include "hw/nmi.h" 54 #include "hw/loader.h" 55 #include "multiboot.h" 56 #include "elf.h" 57 #include "standard-headers/asm-x86/bootparam.h" 58 #include CONFIG_DEVICES 59 #include "kvm/kvm_i386.h" 60 61 #ifdef CONFIG_XEN_EMU 62 #include "hw/xen/xen.h" 63 #include "hw/i386/kvm/xen_evtchn.h" 64 #endif 65 66 /* Physical Address of PVH entry point read from kernel ELF NOTE */ 67 static size_t pvh_start_addr; 68 69 static void init_topo_info(X86CPUTopoInfo *topo_info, 70 const X86MachineState *x86ms) 71 { 72 MachineState *ms = MACHINE(x86ms); 73 74 topo_info->dies_per_pkg = ms->smp.dies; 75 topo_info->cores_per_die = ms->smp.cores; 76 topo_info->threads_per_core = ms->smp.threads; 77 } 78 79 /* 80 * Calculates initial APIC ID for a specific CPU index 81 * 82 * Currently we need to be able to calculate the APIC ID from the CPU index 83 * alone (without requiring a CPU object), as the QEMU<->Seabios interfaces have 84 * no concept of "CPU index", and the NUMA tables on fw_cfg need the APIC ID of 85 * all CPUs up to max_cpus. 86 */ 87 uint32_t x86_cpu_apic_id_from_index(X86MachineState *x86ms, 88 unsigned int cpu_index) 89 { 90 X86CPUTopoInfo topo_info; 91 92 init_topo_info(&topo_info, x86ms); 93 94 return x86_apicid_from_cpu_idx(&topo_info, cpu_index); 95 } 96 97 98 void x86_cpu_new(X86MachineState *x86ms, int64_t apic_id, Error **errp) 99 { 100 Object *cpu = object_new(MACHINE(x86ms)->cpu_type); 101 102 if (!object_property_set_uint(cpu, "apic-id", apic_id, errp)) { 103 goto out; 104 } 105 qdev_realize(DEVICE(cpu), NULL, errp); 106 107 out: 108 object_unref(cpu); 109 } 110 111 void x86_cpus_init(X86MachineState *x86ms, int default_cpu_version) 112 { 113 int i; 114 const CPUArchIdList *possible_cpus; 115 MachineState *ms = MACHINE(x86ms); 116 MachineClass *mc = MACHINE_GET_CLASS(x86ms); 117 118 x86_cpu_set_default_version(default_cpu_version); 119 120 /* 121 * Calculates the limit to CPU APIC ID values 122 * 123 * Limit for the APIC ID value, so that all 124 * CPU APIC IDs are < x86ms->apic_id_limit. 125 * 126 * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create(). 127 */ 128 x86ms->apic_id_limit = x86_cpu_apic_id_from_index(x86ms, 129 ms->smp.max_cpus - 1) + 1; 130 131 /* 132 * Can we support APIC ID 255 or higher? With KVM, that requires 133 * both in-kernel lapic and X2APIC userspace API. 134 * 135 * kvm_enabled() must go first to ensure that kvm_* references are 136 * not emitted for the linker to consume (kvm_enabled() is 137 * a literal `0` in configurations where kvm_* aren't defined) 138 */ 139 if (kvm_enabled() && x86ms->apic_id_limit > 255 && 140 kvm_irqchip_in_kernel() && !kvm_enable_x2apic()) { 141 error_report("current -smp configuration requires kernel " 142 "irqchip and X2APIC API support."); 143 exit(EXIT_FAILURE); 144 } 145 146 if (kvm_enabled()) { 147 kvm_set_max_apic_id(x86ms->apic_id_limit); 148 } 149 150 if (!kvm_irqchip_in_kernel()) { 151 apic_set_max_apic_id(x86ms->apic_id_limit); 152 } 153 154 possible_cpus = mc->possible_cpu_arch_ids(ms); 155 for (i = 0; i < ms->smp.cpus; i++) { 156 x86_cpu_new(x86ms, possible_cpus->cpus[i].arch_id, &error_fatal); 157 } 158 } 159 160 void x86_rtc_set_cpus_count(ISADevice *s, uint16_t cpus_count) 161 { 162 MC146818RtcState *rtc = MC146818_RTC(s); 163 164 if (cpus_count > 0xff) { 165 /* 166 * If the number of CPUs can't be represented in 8 bits, the 167 * BIOS must use "FW_CFG_NB_CPUS". Set RTC field to 0 just 168 * to make old BIOSes fail more predictably. 169 */ 170 mc146818rtc_set_cmos_data(rtc, 0x5f, 0); 171 } else { 172 mc146818rtc_set_cmos_data(rtc, 0x5f, cpus_count - 1); 173 } 174 } 175 176 static int x86_apic_cmp(const void *a, const void *b) 177 { 178 CPUArchId *apic_a = (CPUArchId *)a; 179 CPUArchId *apic_b = (CPUArchId *)b; 180 181 return apic_a->arch_id - apic_b->arch_id; 182 } 183 184 /* 185 * returns pointer to CPUArchId descriptor that matches CPU's apic_id 186 * in ms->possible_cpus->cpus, if ms->possible_cpus->cpus has no 187 * entry corresponding to CPU's apic_id returns NULL. 188 */ 189 CPUArchId *x86_find_cpu_slot(MachineState *ms, uint32_t id, int *idx) 190 { 191 CPUArchId apic_id, *found_cpu; 192 193 apic_id.arch_id = id; 194 found_cpu = bsearch(&apic_id, ms->possible_cpus->cpus, 195 ms->possible_cpus->len, sizeof(*ms->possible_cpus->cpus), 196 x86_apic_cmp); 197 if (found_cpu && idx) { 198 *idx = found_cpu - ms->possible_cpus->cpus; 199 } 200 return found_cpu; 201 } 202 203 void x86_cpu_plug(HotplugHandler *hotplug_dev, 204 DeviceState *dev, Error **errp) 205 { 206 CPUArchId *found_cpu; 207 Error *local_err = NULL; 208 X86CPU *cpu = X86_CPU(dev); 209 X86MachineState *x86ms = X86_MACHINE(hotplug_dev); 210 211 if (x86ms->acpi_dev) { 212 hotplug_handler_plug(x86ms->acpi_dev, dev, &local_err); 213 if (local_err) { 214 goto out; 215 } 216 } 217 218 /* increment the number of CPUs */ 219 x86ms->boot_cpus++; 220 if (x86ms->rtc) { 221 x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus); 222 } 223 if (x86ms->fw_cfg) { 224 fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus); 225 } 226 227 found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL); 228 found_cpu->cpu = OBJECT(dev); 229 out: 230 error_propagate(errp, local_err); 231 } 232 233 void x86_cpu_unplug_request_cb(HotplugHandler *hotplug_dev, 234 DeviceState *dev, Error **errp) 235 { 236 int idx = -1; 237 X86CPU *cpu = X86_CPU(dev); 238 X86MachineState *x86ms = X86_MACHINE(hotplug_dev); 239 240 if (!x86ms->acpi_dev) { 241 error_setg(errp, "CPU hot unplug not supported without ACPI"); 242 return; 243 } 244 245 x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx); 246 assert(idx != -1); 247 if (idx == 0) { 248 error_setg(errp, "Boot CPU is unpluggable"); 249 return; 250 } 251 252 hotplug_handler_unplug_request(x86ms->acpi_dev, dev, 253 errp); 254 } 255 256 void x86_cpu_unplug_cb(HotplugHandler *hotplug_dev, 257 DeviceState *dev, Error **errp) 258 { 259 CPUArchId *found_cpu; 260 Error *local_err = NULL; 261 X86CPU *cpu = X86_CPU(dev); 262 X86MachineState *x86ms = X86_MACHINE(hotplug_dev); 263 264 hotplug_handler_unplug(x86ms->acpi_dev, dev, &local_err); 265 if (local_err) { 266 goto out; 267 } 268 269 found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL); 270 found_cpu->cpu = NULL; 271 qdev_unrealize(dev); 272 273 /* decrement the number of CPUs */ 274 x86ms->boot_cpus--; 275 /* Update the number of CPUs in CMOS */ 276 x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus); 277 fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus); 278 out: 279 error_propagate(errp, local_err); 280 } 281 282 void x86_cpu_pre_plug(HotplugHandler *hotplug_dev, 283 DeviceState *dev, Error **errp) 284 { 285 int idx; 286 CPUState *cs; 287 CPUArchId *cpu_slot; 288 X86CPUTopoIDs topo_ids; 289 X86CPU *cpu = X86_CPU(dev); 290 CPUX86State *env = &cpu->env; 291 MachineState *ms = MACHINE(hotplug_dev); 292 X86MachineState *x86ms = X86_MACHINE(hotplug_dev); 293 unsigned int smp_cores = ms->smp.cores; 294 unsigned int smp_threads = ms->smp.threads; 295 X86CPUTopoInfo topo_info; 296 297 if (!object_dynamic_cast(OBJECT(cpu), ms->cpu_type)) { 298 error_setg(errp, "Invalid CPU type, expected cpu type: '%s'", 299 ms->cpu_type); 300 return; 301 } 302 303 if (x86ms->acpi_dev) { 304 Error *local_err = NULL; 305 306 hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms->acpi_dev), dev, 307 &local_err); 308 if (local_err) { 309 error_propagate(errp, local_err); 310 return; 311 } 312 } 313 314 init_topo_info(&topo_info, x86ms); 315 316 env->nr_dies = ms->smp.dies; 317 318 /* 319 * If APIC ID is not set, 320 * set it based on socket/die/core/thread properties. 321 */ 322 if (cpu->apic_id == UNASSIGNED_APIC_ID) { 323 int max_socket = (ms->smp.max_cpus - 1) / 324 smp_threads / smp_cores / ms->smp.dies; 325 326 /* 327 * die-id was optional in QEMU 4.0 and older, so keep it optional 328 * if there's only one die per socket. 329 */ 330 if (cpu->die_id < 0 && ms->smp.dies == 1) { 331 cpu->die_id = 0; 332 } 333 334 if (cpu->socket_id < 0) { 335 error_setg(errp, "CPU socket-id is not set"); 336 return; 337 } else if (cpu->socket_id > max_socket) { 338 error_setg(errp, "Invalid CPU socket-id: %u must be in range 0:%u", 339 cpu->socket_id, max_socket); 340 return; 341 } 342 if (cpu->die_id < 0) { 343 error_setg(errp, "CPU die-id is not set"); 344 return; 345 } else if (cpu->die_id > ms->smp.dies - 1) { 346 error_setg(errp, "Invalid CPU die-id: %u must be in range 0:%u", 347 cpu->die_id, ms->smp.dies - 1); 348 return; 349 } 350 if (cpu->core_id < 0) { 351 error_setg(errp, "CPU core-id is not set"); 352 return; 353 } else if (cpu->core_id > (smp_cores - 1)) { 354 error_setg(errp, "Invalid CPU core-id: %u must be in range 0:%u", 355 cpu->core_id, smp_cores - 1); 356 return; 357 } 358 if (cpu->thread_id < 0) { 359 error_setg(errp, "CPU thread-id is not set"); 360 return; 361 } else if (cpu->thread_id > (smp_threads - 1)) { 362 error_setg(errp, "Invalid CPU thread-id: %u must be in range 0:%u", 363 cpu->thread_id, smp_threads - 1); 364 return; 365 } 366 367 topo_ids.pkg_id = cpu->socket_id; 368 topo_ids.die_id = cpu->die_id; 369 topo_ids.core_id = cpu->core_id; 370 topo_ids.smt_id = cpu->thread_id; 371 cpu->apic_id = x86_apicid_from_topo_ids(&topo_info, &topo_ids); 372 } 373 374 cpu_slot = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx); 375 if (!cpu_slot) { 376 x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids); 377 error_setg(errp, 378 "Invalid CPU [socket: %u, die: %u, core: %u, thread: %u] with" 379 " APIC ID %" PRIu32 ", valid index range 0:%d", 380 topo_ids.pkg_id, topo_ids.die_id, topo_ids.core_id, topo_ids.smt_id, 381 cpu->apic_id, ms->possible_cpus->len - 1); 382 return; 383 } 384 385 if (cpu_slot->cpu) { 386 error_setg(errp, "CPU[%d] with APIC ID %" PRIu32 " exists", 387 idx, cpu->apic_id); 388 return; 389 } 390 391 /* if 'address' properties socket-id/core-id/thread-id are not set, set them 392 * so that machine_query_hotpluggable_cpus would show correct values 393 */ 394 /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn() 395 * once -smp refactoring is complete and there will be CPU private 396 * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */ 397 x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids); 398 if (cpu->socket_id != -1 && cpu->socket_id != topo_ids.pkg_id) { 399 error_setg(errp, "property socket-id: %u doesn't match set apic-id:" 400 " 0x%x (socket-id: %u)", cpu->socket_id, cpu->apic_id, 401 topo_ids.pkg_id); 402 return; 403 } 404 cpu->socket_id = topo_ids.pkg_id; 405 406 if (cpu->die_id != -1 && cpu->die_id != topo_ids.die_id) { 407 error_setg(errp, "property die-id: %u doesn't match set apic-id:" 408 " 0x%x (die-id: %u)", cpu->die_id, cpu->apic_id, topo_ids.die_id); 409 return; 410 } 411 cpu->die_id = topo_ids.die_id; 412 413 if (cpu->core_id != -1 && cpu->core_id != topo_ids.core_id) { 414 error_setg(errp, "property core-id: %u doesn't match set apic-id:" 415 " 0x%x (core-id: %u)", cpu->core_id, cpu->apic_id, 416 topo_ids.core_id); 417 return; 418 } 419 cpu->core_id = topo_ids.core_id; 420 421 if (cpu->thread_id != -1 && cpu->thread_id != topo_ids.smt_id) { 422 error_setg(errp, "property thread-id: %u doesn't match set apic-id:" 423 " 0x%x (thread-id: %u)", cpu->thread_id, cpu->apic_id, 424 topo_ids.smt_id); 425 return; 426 } 427 cpu->thread_id = topo_ids.smt_id; 428 429 /* 430 * kvm_enabled() must go first to ensure that kvm_* references are 431 * not emitted for the linker to consume (kvm_enabled() is 432 * a literal `0` in configurations where kvm_* aren't defined) 433 */ 434 if (kvm_enabled() && hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) && 435 !kvm_hv_vpindex_settable()) { 436 error_setg(errp, "kernel doesn't allow setting HyperV VP_INDEX"); 437 return; 438 } 439 440 cs = CPU(cpu); 441 cs->cpu_index = idx; 442 443 numa_cpu_pre_plug(cpu_slot, dev, errp); 444 } 445 446 CpuInstanceProperties 447 x86_cpu_index_to_props(MachineState *ms, unsigned cpu_index) 448 { 449 MachineClass *mc = MACHINE_GET_CLASS(ms); 450 const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms); 451 452 assert(cpu_index < possible_cpus->len); 453 return possible_cpus->cpus[cpu_index].props; 454 } 455 456 int64_t x86_get_default_cpu_node_id(const MachineState *ms, int idx) 457 { 458 X86CPUTopoIDs topo_ids; 459 X86MachineState *x86ms = X86_MACHINE(ms); 460 X86CPUTopoInfo topo_info; 461 462 init_topo_info(&topo_info, x86ms); 463 464 assert(idx < ms->possible_cpus->len); 465 x86_topo_ids_from_apicid(ms->possible_cpus->cpus[idx].arch_id, 466 &topo_info, &topo_ids); 467 return topo_ids.pkg_id % ms->numa_state->num_nodes; 468 } 469 470 const CPUArchIdList *x86_possible_cpu_arch_ids(MachineState *ms) 471 { 472 X86MachineState *x86ms = X86_MACHINE(ms); 473 unsigned int max_cpus = ms->smp.max_cpus; 474 X86CPUTopoInfo topo_info; 475 int i; 476 477 if (ms->possible_cpus) { 478 /* 479 * make sure that max_cpus hasn't changed since the first use, i.e. 480 * -smp hasn't been parsed after it 481 */ 482 assert(ms->possible_cpus->len == max_cpus); 483 return ms->possible_cpus; 484 } 485 486 ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) + 487 sizeof(CPUArchId) * max_cpus); 488 ms->possible_cpus->len = max_cpus; 489 490 init_topo_info(&topo_info, x86ms); 491 492 for (i = 0; i < ms->possible_cpus->len; i++) { 493 X86CPUTopoIDs topo_ids; 494 495 ms->possible_cpus->cpus[i].type = ms->cpu_type; 496 ms->possible_cpus->cpus[i].vcpus_count = 1; 497 ms->possible_cpus->cpus[i].arch_id = 498 x86_cpu_apic_id_from_index(x86ms, i); 499 x86_topo_ids_from_apicid(ms->possible_cpus->cpus[i].arch_id, 500 &topo_info, &topo_ids); 501 ms->possible_cpus->cpus[i].props.has_socket_id = true; 502 ms->possible_cpus->cpus[i].props.socket_id = topo_ids.pkg_id; 503 if (ms->smp.dies > 1) { 504 ms->possible_cpus->cpus[i].props.has_die_id = true; 505 ms->possible_cpus->cpus[i].props.die_id = topo_ids.die_id; 506 } 507 ms->possible_cpus->cpus[i].props.has_core_id = true; 508 ms->possible_cpus->cpus[i].props.core_id = topo_ids.core_id; 509 ms->possible_cpus->cpus[i].props.has_thread_id = true; 510 ms->possible_cpus->cpus[i].props.thread_id = topo_ids.smt_id; 511 } 512 return ms->possible_cpus; 513 } 514 515 static void x86_nmi(NMIState *n, int cpu_index, Error **errp) 516 { 517 /* cpu index isn't used */ 518 CPUState *cs; 519 520 CPU_FOREACH(cs) { 521 X86CPU *cpu = X86_CPU(cs); 522 523 if (cpu_is_apic_enabled(cpu->apic_state)) { 524 apic_deliver_nmi(cpu->apic_state); 525 } else { 526 cpu_interrupt(cs, CPU_INTERRUPT_NMI); 527 } 528 } 529 } 530 531 static long get_file_size(FILE *f) 532 { 533 long where, size; 534 535 /* XXX: on Unix systems, using fstat() probably makes more sense */ 536 537 where = ftell(f); 538 fseek(f, 0, SEEK_END); 539 size = ftell(f); 540 fseek(f, where, SEEK_SET); 541 542 return size; 543 } 544 545 /* TSC handling */ 546 uint64_t cpu_get_tsc(CPUX86State *env) 547 { 548 return cpus_get_elapsed_ticks(); 549 } 550 551 /* IRQ handling */ 552 static void pic_irq_request(void *opaque, int irq, int level) 553 { 554 CPUState *cs = first_cpu; 555 X86CPU *cpu = X86_CPU(cs); 556 557 trace_x86_pic_interrupt(irq, level); 558 if (cpu_is_apic_enabled(cpu->apic_state) && !kvm_irqchip_in_kernel() && 559 !whpx_apic_in_platform()) { 560 CPU_FOREACH(cs) { 561 cpu = X86_CPU(cs); 562 if (apic_accept_pic_intr(cpu->apic_state)) { 563 apic_deliver_pic_intr(cpu->apic_state, level); 564 } 565 } 566 } else { 567 if (level) { 568 cpu_interrupt(cs, CPU_INTERRUPT_HARD); 569 } else { 570 cpu_reset_interrupt(cs, CPU_INTERRUPT_HARD); 571 } 572 } 573 } 574 575 qemu_irq x86_allocate_cpu_irq(void) 576 { 577 return qemu_allocate_irq(pic_irq_request, NULL, 0); 578 } 579 580 int cpu_get_pic_interrupt(CPUX86State *env) 581 { 582 X86CPU *cpu = env_archcpu(env); 583 int intno; 584 585 if (!kvm_irqchip_in_kernel() && !whpx_apic_in_platform()) { 586 intno = apic_get_interrupt(cpu->apic_state); 587 if (intno >= 0) { 588 return intno; 589 } 590 /* read the irq from the PIC */ 591 if (!apic_accept_pic_intr(cpu->apic_state)) { 592 return -1; 593 } 594 } 595 596 intno = pic_read_irq(isa_pic); 597 return intno; 598 } 599 600 DeviceState *cpu_get_current_apic(void) 601 { 602 if (current_cpu) { 603 X86CPU *cpu = X86_CPU(current_cpu); 604 return cpu->apic_state; 605 } else { 606 return NULL; 607 } 608 } 609 610 void gsi_handler(void *opaque, int n, int level) 611 { 612 GSIState *s = opaque; 613 614 trace_x86_gsi_interrupt(n, level); 615 switch (n) { 616 case 0 ... ISA_NUM_IRQS - 1: 617 if (s->i8259_irq[n]) { 618 /* Under KVM, Kernel will forward to both PIC and IOAPIC */ 619 qemu_set_irq(s->i8259_irq[n], level); 620 } 621 /* fall through */ 622 case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1: 623 #ifdef CONFIG_XEN_EMU 624 /* 625 * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC 626 * routing actually works properly under Xen). And then to 627 * *either* the PIRQ handling or the I/OAPIC depending on 628 * whether the former wants it. 629 */ 630 if (xen_mode == XEN_EMULATE && xen_evtchn_set_gsi(n, level)) { 631 break; 632 } 633 #endif 634 qemu_set_irq(s->ioapic_irq[n], level); 635 break; 636 case IO_APIC_SECONDARY_IRQBASE 637 ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1: 638 qemu_set_irq(s->ioapic2_irq[n - IO_APIC_SECONDARY_IRQBASE], level); 639 break; 640 } 641 } 642 643 void ioapic_init_gsi(GSIState *gsi_state, const char *parent_name) 644 { 645 DeviceState *dev; 646 SysBusDevice *d; 647 unsigned int i; 648 649 assert(parent_name); 650 if (kvm_ioapic_in_kernel()) { 651 dev = qdev_new(TYPE_KVM_IOAPIC); 652 } else { 653 dev = qdev_new(TYPE_IOAPIC); 654 } 655 object_property_add_child(object_resolve_path(parent_name, NULL), 656 "ioapic", OBJECT(dev)); 657 d = SYS_BUS_DEVICE(dev); 658 sysbus_realize_and_unref(d, &error_fatal); 659 sysbus_mmio_map(d, 0, IO_APIC_DEFAULT_ADDRESS); 660 661 for (i = 0; i < IOAPIC_NUM_PINS; i++) { 662 gsi_state->ioapic_irq[i] = qdev_get_gpio_in(dev, i); 663 } 664 } 665 666 DeviceState *ioapic_init_secondary(GSIState *gsi_state) 667 { 668 DeviceState *dev; 669 SysBusDevice *d; 670 unsigned int i; 671 672 dev = qdev_new(TYPE_IOAPIC); 673 d = SYS_BUS_DEVICE(dev); 674 sysbus_realize_and_unref(d, &error_fatal); 675 sysbus_mmio_map(d, 0, IO_APIC_SECONDARY_ADDRESS); 676 677 for (i = 0; i < IOAPIC_NUM_PINS; i++) { 678 gsi_state->ioapic2_irq[i] = qdev_get_gpio_in(dev, i); 679 } 680 return dev; 681 } 682 683 struct setup_data { 684 uint64_t next; 685 uint32_t type; 686 uint32_t len; 687 uint8_t data[]; 688 } __attribute__((packed)); 689 690 691 /* 692 * The entry point into the kernel for PVH boot is different from 693 * the native entry point. The PVH entry is defined by the x86/HVM 694 * direct boot ABI and is available in an ELFNOTE in the kernel binary. 695 * 696 * This function is passed to load_elf() when it is called from 697 * load_elfboot() which then additionally checks for an ELF Note of 698 * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to 699 * parse the PVH entry address from the ELF Note. 700 * 701 * Due to trickery in elf_opts.h, load_elf() is actually available as 702 * load_elf32() or load_elf64() and this routine needs to be able 703 * to deal with being called as 32 or 64 bit. 704 * 705 * The address of the PVH entry point is saved to the 'pvh_start_addr' 706 * global variable. (although the entry point is 32-bit, the kernel 707 * binary can be either 32-bit or 64-bit). 708 */ 709 static uint64_t read_pvh_start_addr(void *arg1, void *arg2, bool is64) 710 { 711 size_t *elf_note_data_addr; 712 713 /* Check if ELF Note header passed in is valid */ 714 if (arg1 == NULL) { 715 return 0; 716 } 717 718 if (is64) { 719 struct elf64_note *nhdr64 = (struct elf64_note *)arg1; 720 uint64_t nhdr_size64 = sizeof(struct elf64_note); 721 uint64_t phdr_align = *(uint64_t *)arg2; 722 uint64_t nhdr_namesz = nhdr64->n_namesz; 723 724 elf_note_data_addr = 725 ((void *)nhdr64) + nhdr_size64 + 726 QEMU_ALIGN_UP(nhdr_namesz, phdr_align); 727 728 pvh_start_addr = *elf_note_data_addr; 729 } else { 730 struct elf32_note *nhdr32 = (struct elf32_note *)arg1; 731 uint32_t nhdr_size32 = sizeof(struct elf32_note); 732 uint32_t phdr_align = *(uint32_t *)arg2; 733 uint32_t nhdr_namesz = nhdr32->n_namesz; 734 735 elf_note_data_addr = 736 ((void *)nhdr32) + nhdr_size32 + 737 QEMU_ALIGN_UP(nhdr_namesz, phdr_align); 738 739 pvh_start_addr = *(uint32_t *)elf_note_data_addr; 740 } 741 742 return pvh_start_addr; 743 } 744 745 static bool load_elfboot(const char *kernel_filename, 746 int kernel_file_size, 747 uint8_t *header, 748 size_t pvh_xen_start_addr, 749 FWCfgState *fw_cfg) 750 { 751 uint32_t flags = 0; 752 uint32_t mh_load_addr = 0; 753 uint32_t elf_kernel_size = 0; 754 uint64_t elf_entry; 755 uint64_t elf_low, elf_high; 756 int kernel_size; 757 758 if (ldl_p(header) != 0x464c457f) { 759 return false; /* no elfboot */ 760 } 761 762 bool elf_is64 = header[EI_CLASS] == ELFCLASS64; 763 flags = elf_is64 ? 764 ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags; 765 766 if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */ 767 error_report("elfboot unsupported flags = %x", flags); 768 exit(1); 769 } 770 771 uint64_t elf_note_type = XEN_ELFNOTE_PHYS32_ENTRY; 772 kernel_size = load_elf(kernel_filename, read_pvh_start_addr, 773 NULL, &elf_note_type, &elf_entry, 774 &elf_low, &elf_high, NULL, 0, I386_ELF_MACHINE, 775 0, 0); 776 777 if (kernel_size < 0) { 778 error_report("Error while loading elf kernel"); 779 exit(1); 780 } 781 mh_load_addr = elf_low; 782 elf_kernel_size = elf_high - elf_low; 783 784 if (pvh_start_addr == 0) { 785 error_report("Error loading uncompressed kernel without PVH ELF Note"); 786 exit(1); 787 } 788 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_start_addr); 789 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr); 790 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size); 791 792 return true; 793 } 794 795 void x86_load_linux(X86MachineState *x86ms, 796 FWCfgState *fw_cfg, 797 int acpi_data_size, 798 bool pvh_enabled) 799 { 800 bool linuxboot_dma_enabled = X86_MACHINE_GET_CLASS(x86ms)->fwcfg_dma_enabled; 801 uint16_t protocol; 802 int setup_size, kernel_size, cmdline_size; 803 int dtb_size, setup_data_offset; 804 uint32_t initrd_max; 805 uint8_t header[8192], *setup, *kernel; 806 hwaddr real_addr, prot_addr, cmdline_addr, initrd_addr = 0; 807 FILE *f; 808 char *vmode; 809 MachineState *machine = MACHINE(x86ms); 810 struct setup_data *setup_data; 811 const char *kernel_filename = machine->kernel_filename; 812 const char *initrd_filename = machine->initrd_filename; 813 const char *dtb_filename = machine->dtb; 814 const char *kernel_cmdline = machine->kernel_cmdline; 815 SevKernelLoaderContext sev_load_ctx = {}; 816 817 /* Align to 16 bytes as a paranoia measure */ 818 cmdline_size = (strlen(kernel_cmdline) + 16) & ~15; 819 820 /* load the kernel header */ 821 f = fopen(kernel_filename, "rb"); 822 if (!f) { 823 fprintf(stderr, "qemu: could not open kernel file '%s': %s\n", 824 kernel_filename, strerror(errno)); 825 exit(1); 826 } 827 828 kernel_size = get_file_size(f); 829 if (!kernel_size || 830 fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) != 831 MIN(ARRAY_SIZE(header), kernel_size)) { 832 fprintf(stderr, "qemu: could not load kernel '%s': %s\n", 833 kernel_filename, strerror(errno)); 834 exit(1); 835 } 836 837 /* kernel protocol version */ 838 if (ldl_p(header + 0x202) == 0x53726448) { 839 protocol = lduw_p(header + 0x206); 840 } else { 841 /* 842 * This could be a multiboot kernel. If it is, let's stop treating it 843 * like a Linux kernel. 844 * Note: some multiboot images could be in the ELF format (the same of 845 * PVH), so we try multiboot first since we check the multiboot magic 846 * header before to load it. 847 */ 848 if (load_multiboot(x86ms, fw_cfg, f, kernel_filename, initrd_filename, 849 kernel_cmdline, kernel_size, header)) { 850 return; 851 } 852 /* 853 * Check if the file is an uncompressed kernel file (ELF) and load it, 854 * saving the PVH entry point used by the x86/HVM direct boot ABI. 855 * If load_elfboot() is successful, populate the fw_cfg info. 856 */ 857 if (pvh_enabled && 858 load_elfboot(kernel_filename, kernel_size, 859 header, pvh_start_addr, fw_cfg)) { 860 fclose(f); 861 862 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, 863 strlen(kernel_cmdline) + 1); 864 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline); 865 866 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header)); 867 fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, 868 header, sizeof(header)); 869 870 /* load initrd */ 871 if (initrd_filename) { 872 GMappedFile *mapped_file; 873 gsize initrd_size; 874 gchar *initrd_data; 875 GError *gerr = NULL; 876 877 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr); 878 if (!mapped_file) { 879 fprintf(stderr, "qemu: error reading initrd %s: %s\n", 880 initrd_filename, gerr->message); 881 exit(1); 882 } 883 x86ms->initrd_mapped_file = mapped_file; 884 885 initrd_data = g_mapped_file_get_contents(mapped_file); 886 initrd_size = g_mapped_file_get_length(mapped_file); 887 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1; 888 if (initrd_size >= initrd_max) { 889 fprintf(stderr, "qemu: initrd is too large, cannot support." 890 "(max: %"PRIu32", need %"PRId64")\n", 891 initrd_max, (uint64_t)initrd_size); 892 exit(1); 893 } 894 895 initrd_addr = (initrd_max - initrd_size) & ~4095; 896 897 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr); 898 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size); 899 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, 900 initrd_size); 901 } 902 903 option_rom[nb_option_roms].bootindex = 0; 904 option_rom[nb_option_roms].name = "pvh.bin"; 905 nb_option_roms++; 906 907 return; 908 } 909 protocol = 0; 910 } 911 912 if (protocol < 0x200 || !(header[0x211] & 0x01)) { 913 /* Low kernel */ 914 real_addr = 0x90000; 915 cmdline_addr = 0x9a000 - cmdline_size; 916 prot_addr = 0x10000; 917 } else if (protocol < 0x202) { 918 /* High but ancient kernel */ 919 real_addr = 0x90000; 920 cmdline_addr = 0x9a000 - cmdline_size; 921 prot_addr = 0x100000; 922 } else { 923 /* High and recent kernel */ 924 real_addr = 0x10000; 925 cmdline_addr = 0x20000; 926 prot_addr = 0x100000; 927 } 928 929 /* highest address for loading the initrd */ 930 if (protocol >= 0x20c && 931 lduw_p(header + 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G) { 932 /* 933 * Linux has supported initrd up to 4 GB for a very long time (2007, 934 * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013), 935 * though it only sets initrd_max to 2 GB to "work around bootloader 936 * bugs". Luckily, QEMU firmware(which does something like bootloader) 937 * has supported this. 938 * 939 * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can 940 * be loaded into any address. 941 * 942 * In addition, initrd_max is uint32_t simply because QEMU doesn't 943 * support the 64-bit boot protocol (specifically the ext_ramdisk_image 944 * field). 945 * 946 * Therefore here just limit initrd_max to UINT32_MAX simply as well. 947 */ 948 initrd_max = UINT32_MAX; 949 } else if (protocol >= 0x203) { 950 initrd_max = ldl_p(header + 0x22c); 951 } else { 952 initrd_max = 0x37ffffff; 953 } 954 955 if (initrd_max >= x86ms->below_4g_mem_size - acpi_data_size) { 956 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1; 957 } 958 959 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr); 960 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(kernel_cmdline) + 1); 961 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline); 962 sev_load_ctx.cmdline_data = (char *)kernel_cmdline; 963 sev_load_ctx.cmdline_size = strlen(kernel_cmdline) + 1; 964 965 if (protocol >= 0x202) { 966 stl_p(header + 0x228, cmdline_addr); 967 } else { 968 stw_p(header + 0x20, 0xA33F); 969 stw_p(header + 0x22, cmdline_addr - real_addr); 970 } 971 972 /* handle vga= parameter */ 973 vmode = strstr(kernel_cmdline, "vga="); 974 if (vmode) { 975 unsigned int video_mode; 976 const char *end; 977 int ret; 978 /* skip "vga=" */ 979 vmode += 4; 980 if (!strncmp(vmode, "normal", 6)) { 981 video_mode = 0xffff; 982 } else if (!strncmp(vmode, "ext", 3)) { 983 video_mode = 0xfffe; 984 } else if (!strncmp(vmode, "ask", 3)) { 985 video_mode = 0xfffd; 986 } else { 987 ret = qemu_strtoui(vmode, &end, 0, &video_mode); 988 if (ret != 0 || (*end && *end != ' ')) { 989 fprintf(stderr, "qemu: invalid 'vga=' kernel parameter.\n"); 990 exit(1); 991 } 992 } 993 stw_p(header + 0x1fa, video_mode); 994 } 995 996 /* loader type */ 997 /* 998 * High nybble = B reserved for QEMU; low nybble is revision number. 999 * If this code is substantially changed, you may want to consider 1000 * incrementing the revision. 1001 */ 1002 if (protocol >= 0x200) { 1003 header[0x210] = 0xB0; 1004 } 1005 /* heap */ 1006 if (protocol >= 0x201) { 1007 header[0x211] |= 0x80; /* CAN_USE_HEAP */ 1008 stw_p(header + 0x224, cmdline_addr - real_addr - 0x200); 1009 } 1010 1011 /* load initrd */ 1012 if (initrd_filename) { 1013 GMappedFile *mapped_file; 1014 gsize initrd_size; 1015 gchar *initrd_data; 1016 GError *gerr = NULL; 1017 1018 if (protocol < 0x200) { 1019 fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n"); 1020 exit(1); 1021 } 1022 1023 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr); 1024 if (!mapped_file) { 1025 fprintf(stderr, "qemu: error reading initrd %s: %s\n", 1026 initrd_filename, gerr->message); 1027 exit(1); 1028 } 1029 x86ms->initrd_mapped_file = mapped_file; 1030 1031 initrd_data = g_mapped_file_get_contents(mapped_file); 1032 initrd_size = g_mapped_file_get_length(mapped_file); 1033 if (initrd_size >= initrd_max) { 1034 fprintf(stderr, "qemu: initrd is too large, cannot support." 1035 "(max: %"PRIu32", need %"PRId64")\n", 1036 initrd_max, (uint64_t)initrd_size); 1037 exit(1); 1038 } 1039 1040 initrd_addr = (initrd_max - initrd_size) & ~4095; 1041 1042 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr); 1043 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size); 1044 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size); 1045 sev_load_ctx.initrd_data = initrd_data; 1046 sev_load_ctx.initrd_size = initrd_size; 1047 1048 stl_p(header + 0x218, initrd_addr); 1049 stl_p(header + 0x21c, initrd_size); 1050 } 1051 1052 /* load kernel and setup */ 1053 setup_size = header[0x1f1]; 1054 if (setup_size == 0) { 1055 setup_size = 4; 1056 } 1057 setup_size = (setup_size + 1) * 512; 1058 if (setup_size > kernel_size) { 1059 fprintf(stderr, "qemu: invalid kernel header\n"); 1060 exit(1); 1061 } 1062 kernel_size -= setup_size; 1063 1064 setup = g_malloc(setup_size); 1065 kernel = g_malloc(kernel_size); 1066 fseek(f, 0, SEEK_SET); 1067 if (fread(setup, 1, setup_size, f) != setup_size) { 1068 fprintf(stderr, "fread() failed\n"); 1069 exit(1); 1070 } 1071 if (fread(kernel, 1, kernel_size, f) != kernel_size) { 1072 fprintf(stderr, "fread() failed\n"); 1073 exit(1); 1074 } 1075 fclose(f); 1076 1077 /* append dtb to kernel */ 1078 if (dtb_filename) { 1079 if (protocol < 0x209) { 1080 fprintf(stderr, "qemu: Linux kernel too old to load a dtb\n"); 1081 exit(1); 1082 } 1083 1084 dtb_size = get_image_size(dtb_filename); 1085 if (dtb_size <= 0) { 1086 fprintf(stderr, "qemu: error reading dtb %s: %s\n", 1087 dtb_filename, strerror(errno)); 1088 exit(1); 1089 } 1090 1091 setup_data_offset = QEMU_ALIGN_UP(kernel_size, 16); 1092 kernel_size = setup_data_offset + sizeof(struct setup_data) + dtb_size; 1093 kernel = g_realloc(kernel, kernel_size); 1094 1095 stq_p(header + 0x250, prot_addr + setup_data_offset); 1096 1097 setup_data = (struct setup_data *)(kernel + setup_data_offset); 1098 setup_data->next = 0; 1099 setup_data->type = cpu_to_le32(SETUP_DTB); 1100 setup_data->len = cpu_to_le32(dtb_size); 1101 1102 load_image_size(dtb_filename, setup_data->data, dtb_size); 1103 } 1104 1105 /* 1106 * If we're starting an encrypted VM, it will be OVMF based, which uses the 1107 * efi stub for booting and doesn't require any values to be placed in the 1108 * kernel header. We therefore don't update the header so the hash of the 1109 * kernel on the other side of the fw_cfg interface matches the hash of the 1110 * file the user passed in. 1111 */ 1112 if (!sev_enabled()) { 1113 memcpy(setup, header, MIN(sizeof(header), setup_size)); 1114 } 1115 1116 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr); 1117 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size); 1118 fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA, kernel, kernel_size); 1119 sev_load_ctx.kernel_data = (char *)kernel; 1120 sev_load_ctx.kernel_size = kernel_size; 1121 1122 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr); 1123 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size); 1124 fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size); 1125 sev_load_ctx.setup_data = (char *)setup; 1126 sev_load_ctx.setup_size = setup_size; 1127 1128 if (sev_enabled()) { 1129 sev_add_kernel_loader_hashes(&sev_load_ctx, &error_fatal); 1130 } 1131 1132 option_rom[nb_option_roms].bootindex = 0; 1133 option_rom[nb_option_roms].name = "linuxboot.bin"; 1134 if (linuxboot_dma_enabled && fw_cfg_dma_enabled(fw_cfg)) { 1135 option_rom[nb_option_roms].name = "linuxboot_dma.bin"; 1136 } 1137 nb_option_roms++; 1138 } 1139 1140 void x86_bios_rom_init(MachineState *ms, const char *default_firmware, 1141 MemoryRegion *rom_memory, bool isapc_ram_fw) 1142 { 1143 const char *bios_name; 1144 char *filename; 1145 MemoryRegion *bios, *isa_bios; 1146 int bios_size, isa_bios_size; 1147 ssize_t ret; 1148 1149 /* BIOS load */ 1150 bios_name = ms->firmware ?: default_firmware; 1151 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name); 1152 if (filename) { 1153 bios_size = get_image_size(filename); 1154 } else { 1155 bios_size = -1; 1156 } 1157 if (bios_size <= 0 || 1158 (bios_size % 65536) != 0) { 1159 goto bios_error; 1160 } 1161 bios = g_malloc(sizeof(*bios)); 1162 memory_region_init_ram(bios, NULL, "pc.bios", bios_size, &error_fatal); 1163 if (sev_enabled()) { 1164 /* 1165 * The concept of a "reset" simply doesn't exist for 1166 * confidential computing guests, we have to destroy and 1167 * re-launch them instead. So there is no need to register 1168 * the firmware as rom to properly re-initialize on reset. 1169 * Just go for a straight file load instead. 1170 */ 1171 void *ptr = memory_region_get_ram_ptr(bios); 1172 load_image_size(filename, ptr, bios_size); 1173 x86_firmware_configure(ptr, bios_size); 1174 } else { 1175 if (!isapc_ram_fw) { 1176 memory_region_set_readonly(bios, true); 1177 } 1178 ret = rom_add_file_fixed(bios_name, (uint32_t)(-bios_size), -1); 1179 if (ret != 0) { 1180 goto bios_error; 1181 } 1182 } 1183 g_free(filename); 1184 1185 /* map the last 128KB of the BIOS in ISA space */ 1186 isa_bios_size = MIN(bios_size, 128 * KiB); 1187 isa_bios = g_malloc(sizeof(*isa_bios)); 1188 memory_region_init_alias(isa_bios, NULL, "isa-bios", bios, 1189 bios_size - isa_bios_size, isa_bios_size); 1190 memory_region_add_subregion_overlap(rom_memory, 1191 0x100000 - isa_bios_size, 1192 isa_bios, 1193 1); 1194 if (!isapc_ram_fw) { 1195 memory_region_set_readonly(isa_bios, true); 1196 } 1197 1198 /* map all the bios at the top of memory */ 1199 memory_region_add_subregion(rom_memory, 1200 (uint32_t)(-bios_size), 1201 bios); 1202 return; 1203 1204 bios_error: 1205 fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name); 1206 exit(1); 1207 } 1208 1209 bool x86_machine_is_smm_enabled(const X86MachineState *x86ms) 1210 { 1211 bool smm_available = false; 1212 1213 if (x86ms->smm == ON_OFF_AUTO_OFF) { 1214 return false; 1215 } 1216 1217 if (tcg_enabled() || qtest_enabled()) { 1218 smm_available = true; 1219 } else if (kvm_enabled()) { 1220 smm_available = kvm_has_smm(); 1221 } 1222 1223 if (smm_available) { 1224 return true; 1225 } 1226 1227 if (x86ms->smm == ON_OFF_AUTO_ON) { 1228 error_report("System Management Mode not supported by this hypervisor."); 1229 exit(1); 1230 } 1231 return false; 1232 } 1233 1234 static void x86_machine_get_smm(Object *obj, Visitor *v, const char *name, 1235 void *opaque, Error **errp) 1236 { 1237 X86MachineState *x86ms = X86_MACHINE(obj); 1238 OnOffAuto smm = x86ms->smm; 1239 1240 visit_type_OnOffAuto(v, name, &smm, errp); 1241 } 1242 1243 static void x86_machine_set_smm(Object *obj, Visitor *v, const char *name, 1244 void *opaque, Error **errp) 1245 { 1246 X86MachineState *x86ms = X86_MACHINE(obj); 1247 1248 visit_type_OnOffAuto(v, name, &x86ms->smm, errp); 1249 } 1250 1251 bool x86_machine_is_acpi_enabled(const X86MachineState *x86ms) 1252 { 1253 if (x86ms->acpi == ON_OFF_AUTO_OFF) { 1254 return false; 1255 } 1256 return true; 1257 } 1258 1259 static void x86_machine_get_acpi(Object *obj, Visitor *v, const char *name, 1260 void *opaque, Error **errp) 1261 { 1262 X86MachineState *x86ms = X86_MACHINE(obj); 1263 OnOffAuto acpi = x86ms->acpi; 1264 1265 visit_type_OnOffAuto(v, name, &acpi, errp); 1266 } 1267 1268 static void x86_machine_set_acpi(Object *obj, Visitor *v, const char *name, 1269 void *opaque, Error **errp) 1270 { 1271 X86MachineState *x86ms = X86_MACHINE(obj); 1272 1273 visit_type_OnOffAuto(v, name, &x86ms->acpi, errp); 1274 } 1275 1276 static void x86_machine_get_pit(Object *obj, Visitor *v, const char *name, 1277 void *opaque, Error **errp) 1278 { 1279 X86MachineState *x86ms = X86_MACHINE(obj); 1280 OnOffAuto pit = x86ms->pit; 1281 1282 visit_type_OnOffAuto(v, name, &pit, errp); 1283 } 1284 1285 static void x86_machine_set_pit(Object *obj, Visitor *v, const char *name, 1286 void *opaque, Error **errp) 1287 { 1288 X86MachineState *x86ms = X86_MACHINE(obj);; 1289 1290 visit_type_OnOffAuto(v, name, &x86ms->pit, errp); 1291 } 1292 1293 static void x86_machine_get_pic(Object *obj, Visitor *v, const char *name, 1294 void *opaque, Error **errp) 1295 { 1296 X86MachineState *x86ms = X86_MACHINE(obj); 1297 OnOffAuto pic = x86ms->pic; 1298 1299 visit_type_OnOffAuto(v, name, &pic, errp); 1300 } 1301 1302 static void x86_machine_set_pic(Object *obj, Visitor *v, const char *name, 1303 void *opaque, Error **errp) 1304 { 1305 X86MachineState *x86ms = X86_MACHINE(obj); 1306 1307 visit_type_OnOffAuto(v, name, &x86ms->pic, errp); 1308 } 1309 1310 static char *x86_machine_get_oem_id(Object *obj, Error **errp) 1311 { 1312 X86MachineState *x86ms = X86_MACHINE(obj); 1313 1314 return g_strdup(x86ms->oem_id); 1315 } 1316 1317 static void x86_machine_set_oem_id(Object *obj, const char *value, Error **errp) 1318 { 1319 X86MachineState *x86ms = X86_MACHINE(obj); 1320 size_t len = strlen(value); 1321 1322 if (len > 6) { 1323 error_setg(errp, 1324 "User specified "X86_MACHINE_OEM_ID" value is bigger than " 1325 "6 bytes in size"); 1326 return; 1327 } 1328 1329 strncpy(x86ms->oem_id, value, 6); 1330 } 1331 1332 static char *x86_machine_get_oem_table_id(Object *obj, Error **errp) 1333 { 1334 X86MachineState *x86ms = X86_MACHINE(obj); 1335 1336 return g_strdup(x86ms->oem_table_id); 1337 } 1338 1339 static void x86_machine_set_oem_table_id(Object *obj, const char *value, 1340 Error **errp) 1341 { 1342 X86MachineState *x86ms = X86_MACHINE(obj); 1343 size_t len = strlen(value); 1344 1345 if (len > 8) { 1346 error_setg(errp, 1347 "User specified "X86_MACHINE_OEM_TABLE_ID 1348 " value is bigger than " 1349 "8 bytes in size"); 1350 return; 1351 } 1352 strncpy(x86ms->oem_table_id, value, 8); 1353 } 1354 1355 static void x86_machine_get_bus_lock_ratelimit(Object *obj, Visitor *v, 1356 const char *name, void *opaque, Error **errp) 1357 { 1358 X86MachineState *x86ms = X86_MACHINE(obj); 1359 uint64_t bus_lock_ratelimit = x86ms->bus_lock_ratelimit; 1360 1361 visit_type_uint64(v, name, &bus_lock_ratelimit, errp); 1362 } 1363 1364 static void x86_machine_set_bus_lock_ratelimit(Object *obj, Visitor *v, 1365 const char *name, void *opaque, Error **errp) 1366 { 1367 X86MachineState *x86ms = X86_MACHINE(obj); 1368 1369 visit_type_uint64(v, name, &x86ms->bus_lock_ratelimit, errp); 1370 } 1371 1372 static void machine_get_sgx_epc(Object *obj, Visitor *v, const char *name, 1373 void *opaque, Error **errp) 1374 { 1375 X86MachineState *x86ms = X86_MACHINE(obj); 1376 SgxEPCList *list = x86ms->sgx_epc_list; 1377 1378 visit_type_SgxEPCList(v, name, &list, errp); 1379 } 1380 1381 static void machine_set_sgx_epc(Object *obj, Visitor *v, const char *name, 1382 void *opaque, Error **errp) 1383 { 1384 X86MachineState *x86ms = X86_MACHINE(obj); 1385 SgxEPCList *list; 1386 1387 list = x86ms->sgx_epc_list; 1388 visit_type_SgxEPCList(v, name, &x86ms->sgx_epc_list, errp); 1389 1390 qapi_free_SgxEPCList(list); 1391 } 1392 1393 static void x86_machine_initfn(Object *obj) 1394 { 1395 X86MachineState *x86ms = X86_MACHINE(obj); 1396 1397 x86ms->smm = ON_OFF_AUTO_AUTO; 1398 x86ms->acpi = ON_OFF_AUTO_AUTO; 1399 x86ms->pit = ON_OFF_AUTO_AUTO; 1400 x86ms->pic = ON_OFF_AUTO_AUTO; 1401 x86ms->pci_irq_mask = ACPI_BUILD_PCI_IRQS; 1402 x86ms->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6); 1403 x86ms->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8); 1404 x86ms->bus_lock_ratelimit = 0; 1405 x86ms->above_4g_mem_start = 4 * GiB; 1406 } 1407 1408 static void x86_machine_class_init(ObjectClass *oc, void *data) 1409 { 1410 MachineClass *mc = MACHINE_CLASS(oc); 1411 X86MachineClass *x86mc = X86_MACHINE_CLASS(oc); 1412 NMIClass *nc = NMI_CLASS(oc); 1413 1414 mc->cpu_index_to_instance_props = x86_cpu_index_to_props; 1415 mc->get_default_cpu_node_id = x86_get_default_cpu_node_id; 1416 mc->possible_cpu_arch_ids = x86_possible_cpu_arch_ids; 1417 x86mc->save_tsc_khz = true; 1418 x86mc->fwcfg_dma_enabled = true; 1419 nc->nmi_monitor_handler = x86_nmi; 1420 1421 object_class_property_add(oc, X86_MACHINE_SMM, "OnOffAuto", 1422 x86_machine_get_smm, x86_machine_set_smm, 1423 NULL, NULL); 1424 object_class_property_set_description(oc, X86_MACHINE_SMM, 1425 "Enable SMM"); 1426 1427 object_class_property_add(oc, X86_MACHINE_ACPI, "OnOffAuto", 1428 x86_machine_get_acpi, x86_machine_set_acpi, 1429 NULL, NULL); 1430 object_class_property_set_description(oc, X86_MACHINE_ACPI, 1431 "Enable ACPI"); 1432 1433 object_class_property_add(oc, X86_MACHINE_PIT, "OnOffAuto", 1434 x86_machine_get_pit, 1435 x86_machine_set_pit, 1436 NULL, NULL); 1437 object_class_property_set_description(oc, X86_MACHINE_PIT, 1438 "Enable i8254 PIT"); 1439 1440 object_class_property_add(oc, X86_MACHINE_PIC, "OnOffAuto", 1441 x86_machine_get_pic, 1442 x86_machine_set_pic, 1443 NULL, NULL); 1444 object_class_property_set_description(oc, X86_MACHINE_PIC, 1445 "Enable i8259 PIC"); 1446 1447 object_class_property_add_str(oc, X86_MACHINE_OEM_ID, 1448 x86_machine_get_oem_id, 1449 x86_machine_set_oem_id); 1450 object_class_property_set_description(oc, X86_MACHINE_OEM_ID, 1451 "Override the default value of field OEMID " 1452 "in ACPI table header." 1453 "The string may be up to 6 bytes in size"); 1454 1455 1456 object_class_property_add_str(oc, X86_MACHINE_OEM_TABLE_ID, 1457 x86_machine_get_oem_table_id, 1458 x86_machine_set_oem_table_id); 1459 object_class_property_set_description(oc, X86_MACHINE_OEM_TABLE_ID, 1460 "Override the default value of field OEM Table ID " 1461 "in ACPI table header." 1462 "The string may be up to 8 bytes in size"); 1463 1464 object_class_property_add(oc, X86_MACHINE_BUS_LOCK_RATELIMIT, "uint64_t", 1465 x86_machine_get_bus_lock_ratelimit, 1466 x86_machine_set_bus_lock_ratelimit, NULL, NULL); 1467 object_class_property_set_description(oc, X86_MACHINE_BUS_LOCK_RATELIMIT, 1468 "Set the ratelimit for the bus locks acquired in VMs"); 1469 1470 object_class_property_add(oc, "sgx-epc", "SgxEPC", 1471 machine_get_sgx_epc, machine_set_sgx_epc, 1472 NULL, NULL); 1473 object_class_property_set_description(oc, "sgx-epc", 1474 "SGX EPC device"); 1475 } 1476 1477 static const TypeInfo x86_machine_info = { 1478 .name = TYPE_X86_MACHINE, 1479 .parent = TYPE_MACHINE, 1480 .abstract = true, 1481 .instance_size = sizeof(X86MachineState), 1482 .instance_init = x86_machine_initfn, 1483 .class_size = sizeof(X86MachineClass), 1484 .class_init = x86_machine_class_init, 1485 .interfaces = (InterfaceInfo[]) { 1486 { TYPE_NMI }, 1487 { } 1488 }, 1489 }; 1490 1491 static void x86_machine_register_types(void) 1492 { 1493 type_register_static(&x86_machine_info); 1494 } 1495 1496 type_init(x86_machine_register_types) 1497