hw/i386/microvm.c

   1 /*
   2  * Copyright (c) 2018 Intel Corporation
   3  * Copyright (c) 2019 Red Hat, Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2 or later, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along with
  15  * this program.  If not, see <http://www.gnu.org/licenses/>.
  16  */
  17
  18 #include "qemu/osdep.h"
  19 #include "qemu/error-report.h"
  20 #include "qemu/cutils.h"
  21 #include "qemu/units.h"
  22 #include "qapi/error.h"
  23 #include "qapi/visitor.h"
  24 #include "qapi/qapi-visit-common.h"
  25 #include "sysemu/sysemu.h"
  26 #include "sysemu/cpus.h"
  27 #include "sysemu/numa.h"
  28 #include "sysemu/reset.h"
  29
  30 #include "hw/loader.h"
  31 #include "hw/irq.h"
  32 #include "hw/kvm/clock.h"
  33 #include "hw/i386/microvm.h"
  34 #include "hw/i386/x86.h"
  35 #include "target/i386/cpu.h"
  36 #include "hw/intc/i8259.h"
  37 #include "hw/timer/i8254.h"
  38 #include "hw/rtc/mc146818rtc.h"
  39 #include "hw/char/serial.h"
  40 #include "hw/i386/topology.h"
  41 #include "hw/i386/e820_memory_layout.h"
  42 #include "hw/i386/fw_cfg.h"
  43 #include "hw/virtio/virtio-mmio.h"
  44
  45 #include "cpu.h"
  46 #include "elf.h"
  47 #include "kvm_i386.h"
  48 #include "hw/xen/start_info.h"
  49
  50 #define MICROVM_BIOS_FILENAME "bios-microvm.bin"
  51
  52 static void microvm_set_rtc(MicrovmMachineState *mms, ISADevice *s)
  53 {
  54     X86MachineState *x86ms = X86_MACHINE(mms);
  55     int val;
  56
  57     val = MIN(x86ms->below_4g_mem_size / KiB, 640);
  58     rtc_set_memory(s, 0x15, val);
  59     rtc_set_memory(s, 0x16, val >> 8);
  60     /* extended memory (next 64MiB) */
  61     if (x86ms->below_4g_mem_size > 1 * MiB) {
  62         val = (x86ms->below_4g_mem_size - 1 * MiB) / KiB;
  63     } else {
  64         val = 0;
  65     }
  66     if (val > 65535) {
  67         val = 65535;
  68     }
  69     rtc_set_memory(s, 0x17, val);
  70     rtc_set_memory(s, 0x18, val >> 8);
  71     rtc_set_memory(s, 0x30, val);
  72     rtc_set_memory(s, 0x31, val >> 8);
  73     /* memory between 16MiB and 4GiB */
  74     if (x86ms->below_4g_mem_size > 16 * MiB) {
  75         val = (x86ms->below_4g_mem_size - 16 * MiB) / (64 * KiB);
  76     } else {
  77         val = 0;
  78     }
  79     if (val > 65535) {
  80         val = 65535;
  81     }
  82     rtc_set_memory(s, 0x34, val);
  83     rtc_set_memory(s, 0x35, val >> 8);
  84     /* memory above 4GiB */
  85     val = x86ms->above_4g_mem_size / 65536;
  86     rtc_set_memory(s, 0x5b, val);
  87     rtc_set_memory(s, 0x5c, val >> 8);
  88     rtc_set_memory(s, 0x5d, val >> 16);
  89 }
  90
  91 static void microvm_gsi_handler(void *opaque, int n, int level)
  92 {
  93     GSIState *s = opaque;
  94
  95     qemu_set_irq(s->ioapic_irq[n], level);
  96 }
  97
  98 static void microvm_devices_init(MicrovmMachineState *mms)
  99 {
 100     X86MachineState *x86ms = X86_MACHINE(mms);
 101     ISABus *isa_bus;
 102     ISADevice *rtc_state;
 103     GSIState *gsi_state;
 104     int i;
 105
 106     /* Core components */
 107
 108     gsi_state = g_malloc0(sizeof(*gsi_state));
 109     if (mms->pic == ON_OFF_AUTO_ON || mms->pic == ON_OFF_AUTO_AUTO) {
 110         x86ms->gsi = qemu_allocate_irqs(gsi_handler, gsi_state, GSI_NUM_PINS);
 111     } else {
 112         x86ms->gsi = qemu_allocate_irqs(microvm_gsi_handler,
 113                                         gsi_state, GSI_NUM_PINS);
 114     }
 115
 116     isa_bus = isa_bus_new(NULL, get_system_memory(), get_system_io(),
 117                           &error_abort);
 118     isa_bus_irqs(isa_bus, x86ms->gsi);
 119
 120     ioapic_init_gsi(gsi_state, "machine");
 121
 122     kvmclock_create();
 123
 124     for (i = 0; i < VIRTIO_NUM_TRANSPORTS; i++) {
 125         sysbus_create_simple("virtio-mmio",
 126                              VIRTIO_MMIO_BASE + i * 512,
 127                              x86ms->gsi[VIRTIO_IRQ_BASE + i]);
 128     }
 129
 130     /* Optional and legacy devices */
 131
 132     if (mms->pic == ON_OFF_AUTO_ON || mms->pic == ON_OFF_AUTO_AUTO) {
 133         qemu_irq *i8259;
 134
 135         i8259 = i8259_init(isa_bus, x86_allocate_cpu_irq());
 136         for (i = 0; i < ISA_NUM_IRQS; i++) {
 137             gsi_state->i8259_irq[i] = i8259[i];
 138         }
 139         g_free(i8259);
 140     }
 141
 142     if (mms->pit == ON_OFF_AUTO_ON || mms->pit == ON_OFF_AUTO_AUTO) {
 143         if (kvm_pit_in_kernel()) {
 144             kvm_pit_init(isa_bus, 0x40);
 145         } else {
 146             i8254_pit_init(isa_bus, 0x40, 0, NULL);
 147         }
 148     }
 149
 150     if (mms->rtc == ON_OFF_AUTO_ON ||
 151         (mms->rtc == ON_OFF_AUTO_AUTO && !kvm_enabled())) {
 152         rtc_state = mc146818_rtc_init(isa_bus, 2000, NULL);
 153         microvm_set_rtc(mms, rtc_state);
 154     }
 155
 156     if (mms->isa_serial) {
 157         serial_hds_isa_init(isa_bus, 0, 1);
 158     }
 159
 160     if (bios_name == NULL) {
 161         bios_name = MICROVM_BIOS_FILENAME;
 162     }
 163     x86_bios_rom_init(get_system_memory(), true);
 164 }
 165
 166 static void microvm_memory_init(MicrovmMachineState *mms)
 167 {
 168     MachineState *machine = MACHINE(mms);
 169     X86MachineState *x86ms = X86_MACHINE(mms);
 170     MemoryRegion *ram, *ram_below_4g, *ram_above_4g;
 171     MemoryRegion *system_memory = get_system_memory();
 172     FWCfgState *fw_cfg;
 173     ram_addr_t lowmem;
 174     int i;
 175
 176     /*
 177      * Check whether RAM fits below 4G (leaving 1/2 GByte for IO memory
 178      * and 256 Mbytes for PCI Express Enhanced Configuration Access Mapping
 179      * also known as MMCFG).
 180      * If it doesn't, we need to split it in chunks below and above 4G.
 181      * In any case, try to make sure that guest addresses aligned at
 182      * 1G boundaries get mapped to host addresses aligned at 1G boundaries.
 183      */
 184     if (machine->ram_size >= 0xb0000000) {
 185         lowmem = 0x80000000;
 186     } else {
 187         lowmem = 0xb0000000;
 188     }
 189
 190     /*
 191      * Handle the machine opt max-ram-below-4g.  It is basically doing
 192      * min(qemu limit, user limit).
 193      */
 194     if (!x86ms->max_ram_below_4g) {
 195         x86ms->max_ram_below_4g = 4 * GiB;
 196     }
 197     if (lowmem > x86ms->max_ram_below_4g) {
 198         lowmem = x86ms->max_ram_below_4g;
 199         if (machine->ram_size - lowmem > lowmem &&
 200             lowmem & (1 * GiB - 1)) {
 201             warn_report("There is possibly poor performance as the ram size "
 202                         " (0x%" PRIx64 ") is more then twice the size of"
 203                         " max-ram-below-4g (%"PRIu64") and"
 204                         " max-ram-below-4g is not a multiple of 1G.",
 205                         (uint64_t)machine->ram_size, x86ms->max_ram_below_4g);
 206         }
 207     }
 208
 209     if (machine->ram_size > lowmem) {
 210         x86ms->above_4g_mem_size = machine->ram_size - lowmem;
 211         x86ms->below_4g_mem_size = lowmem;
 212     } else {
 213         x86ms->above_4g_mem_size = 0;
 214         x86ms->below_4g_mem_size = machine->ram_size;
 215     }
 216
 217     ram = g_malloc(sizeof(*ram));
 218     memory_region_allocate_system_memory(ram, NULL, "microvm.ram",
 219                                          machine->ram_size);
 220
 221     ram_below_4g = g_malloc(sizeof(*ram_below_4g));
 222     memory_region_init_alias(ram_below_4g, NULL, "ram-below-4g", ram,
 223                              0, x86ms->below_4g_mem_size);
 224     memory_region_add_subregion(system_memory, 0, ram_below_4g);
 225
 226     e820_add_entry(0, x86ms->below_4g_mem_size, E820_RAM);
 227
 228     if (x86ms->above_4g_mem_size > 0) {
 229         ram_above_4g = g_malloc(sizeof(*ram_above_4g));
 230         memory_region_init_alias(ram_above_4g, NULL, "ram-above-4g", ram,
 231                                  x86ms->below_4g_mem_size,
 232                                  x86ms->above_4g_mem_size);
 233         memory_region_add_subregion(system_memory, 0x100000000ULL,
 234                                     ram_above_4g);
 235         e820_add_entry(0x100000000ULL, x86ms->above_4g_mem_size, E820_RAM);
 236     }
 237
 238     fw_cfg = fw_cfg_init_io_dma(FW_CFG_IO_BASE, FW_CFG_IO_BASE + 4,
 239                                 &address_space_memory);
 240
 241     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, machine->smp.cpus);
 242     fw_cfg_add_i16(fw_cfg, FW_CFG_MAX_CPUS, machine->smp.max_cpus);
 243     fw_cfg_add_i64(fw_cfg, FW_CFG_RAM_SIZE, (uint64_t)machine->ram_size);
 244     fw_cfg_add_i32(fw_cfg, FW_CFG_IRQ0_OVERRIDE, kvm_allows_irq0_override());
 245     fw_cfg_add_bytes(fw_cfg, FW_CFG_E820_TABLE,
 246                      &e820_reserve, sizeof(e820_reserve));
 247     fw_cfg_add_file(fw_cfg, "etc/e820", e820_table,
 248                     sizeof(struct e820_entry) * e820_get_num_entries());
 249
 250     rom_set_fw(fw_cfg);
 251
 252     if (machine->kernel_filename != NULL) {
 253         x86_load_linux(x86ms, fw_cfg, 0, true, true);
 254     }
 255
 256     if (mms->option_roms) {
 257         for (i = 0; i < nb_option_roms; i++) {
 258             rom_add_option(option_rom[i].name, option_rom[i].bootindex);
 259         }
 260     }
 261
 262     x86ms->fw_cfg = fw_cfg;
 263     x86ms->ioapic_as = &address_space_memory;
 264 }
 265
 266 static gchar *microvm_get_mmio_cmdline(gchar *name)
 267 {
 268     gchar *cmdline;
 269     gchar *separator;
 270     long int index;
 271     int ret;
 272
 273     separator = g_strrstr(name, ".");
 274     if (!separator) {
 275         return NULL;
 276     }
 277
 278     if (qemu_strtol(separator + 1, NULL, 10, &index) != 0) {
 279         return NULL;
 280     }
 281
 282     cmdline = g_malloc0(VIRTIO_CMDLINE_MAXLEN);
 283     ret = g_snprintf(cmdline, VIRTIO_CMDLINE_MAXLEN,
 284                      " virtio_mmio.device=512@0x%lx:%ld",
 285                      VIRTIO_MMIO_BASE + index * 512,
 286                      VIRTIO_IRQ_BASE + index);
 287     if (ret < 0 || ret >= VIRTIO_CMDLINE_MAXLEN) {
 288         g_free(cmdline);
 289         return NULL;
 290     }
 291
 292     return cmdline;
 293 }
 294
 295 static void microvm_fix_kernel_cmdline(MachineState *machine)
 296 {
 297     X86MachineState *x86ms = X86_MACHINE(machine);
 298     BusState *bus;
 299     BusChild *kid;
 300     char *cmdline;
 301
 302     /*
 303      * Find MMIO transports with attached devices, and add them to the kernel
 304      * command line.
 305      *
 306      * Yes, this is a hack, but one that heavily improves the UX without
 307      * introducing any significant issues.
 308      */
 309     cmdline = g_strdup(machine->kernel_cmdline);
 310     bus = sysbus_get_default();
 311     QTAILQ_FOREACH(kid, &bus->children, sibling) {
 312         DeviceState *dev = kid->child;
 313         ObjectClass *class = object_get_class(OBJECT(dev));
 314
 315         if (class == object_class_by_name(TYPE_VIRTIO_MMIO)) {
 316             VirtIOMMIOProxy *mmio = VIRTIO_MMIO(OBJECT(dev));
 317             VirtioBusState *mmio_virtio_bus = &mmio->bus;
 318             BusState *mmio_bus = &mmio_virtio_bus->parent_obj;
 319
 320             if (!QTAILQ_EMPTY(&mmio_bus->children)) {
 321                 gchar *mmio_cmdline = microvm_get_mmio_cmdline(mmio_bus->name);
 322                 if (mmio_cmdline) {
 323                     char *newcmd = g_strjoin(NULL, cmdline, mmio_cmdline, NULL);
 324                     g_free(mmio_cmdline);
 325                     g_free(cmdline);
 326                     cmdline = newcmd;
 327                 }
 328             }
 329         }
 330     }
 331
 332     fw_cfg_modify_i32(x86ms->fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(cmdline) + 1);
 333     fw_cfg_modify_string(x86ms->fw_cfg, FW_CFG_CMDLINE_DATA, cmdline);
 334
 335     g_free(cmdline);
 336 }
 337
 338 static void microvm_machine_state_init(MachineState *machine)
 339 {
 340     MicrovmMachineState *mms = MICROVM_MACHINE(machine);
 341     X86MachineState *x86ms = X86_MACHINE(machine);
 342     Error *local_err = NULL;
 343
 344     microvm_memory_init(mms);
 345
 346     x86_cpus_init(x86ms, CPU_VERSION_LATEST);
 347     if (local_err) {
 348         error_report_err(local_err);
 349         exit(1);
 350     }
 351
 352     microvm_devices_init(mms);
 353 }
 354
 355 static void microvm_machine_reset(MachineState *machine)
 356 {
 357     MicrovmMachineState *mms = MICROVM_MACHINE(machine);
 358     CPUState *cs;
 359     X86CPU *cpu;
 360
 361     if (machine->kernel_filename != NULL &&
 362         mms->auto_kernel_cmdline && !mms->kernel_cmdline_fixed) {
 363         microvm_fix_kernel_cmdline(machine);
 364         mms->kernel_cmdline_fixed = true;
 365     }
 366
 367     qemu_devices_reset();
 368
 369     CPU_FOREACH(cs) {
 370         cpu = X86_CPU(cs);
 371
 372         if (cpu->apic_state) {
 373             device_reset(cpu->apic_state);
 374         }
 375     }
 376 }
 377
 378 static void microvm_machine_get_pic(Object *obj, Visitor *v, const char *name,
 379                                     void *opaque, Error **errp)
 380 {
 381     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 382     OnOffAuto pic = mms->pic;
 383
 384     visit_type_OnOffAuto(v, name, &pic, errp);
 385 }
 386
 387 static void microvm_machine_set_pic(Object *obj, Visitor *v, const char *name,
 388                                     void *opaque, Error **errp)
 389 {
 390     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 391
 392     visit_type_OnOffAuto(v, name, &mms->pic, errp);
 393 }
 394
 395 static void microvm_machine_get_pit(Object *obj, Visitor *v, const char *name,
 396                                     void *opaque, Error **errp)
 397 {
 398     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 399     OnOffAuto pit = mms->pit;
 400
 401     visit_type_OnOffAuto(v, name, &pit, errp);
 402 }
 403
 404 static void microvm_machine_set_pit(Object *obj, Visitor *v, const char *name,
 405                                     void *opaque, Error **errp)
 406 {
 407     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 408
 409     visit_type_OnOffAuto(v, name, &mms->pit, errp);
 410 }
 411
 412 static void microvm_machine_get_rtc(Object *obj, Visitor *v, const char *name,
 413                                     void *opaque, Error **errp)
 414 {
 415     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 416     OnOffAuto rtc = mms->rtc;
 417
 418     visit_type_OnOffAuto(v, name, &rtc, errp);
 419 }
 420
 421 static void microvm_machine_set_rtc(Object *obj, Visitor *v, const char *name,
 422                                     void *opaque, Error **errp)
 423 {
 424     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 425
 426     visit_type_OnOffAuto(v, name, &mms->rtc, errp);
 427 }
 428
 429 static bool microvm_machine_get_isa_serial(Object *obj, Error **errp)
 430 {
 431     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 432
 433     return mms->isa_serial;
 434 }
 435
 436 static void microvm_machine_set_isa_serial(Object *obj, bool value,
 437                                            Error **errp)
 438 {
 439     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 440
 441     mms->isa_serial = value;
 442 }
 443
 444 static bool microvm_machine_get_option_roms(Object *obj, Error **errp)
 445 {
 446     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 447
 448     return mms->option_roms;
 449 }
 450
 451 static void microvm_machine_set_option_roms(Object *obj, bool value,
 452                                             Error **errp)
 453 {
 454     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 455
 456     mms->option_roms = value;
 457 }
 458
 459 static bool microvm_machine_get_auto_kernel_cmdline(Object *obj, Error **errp)
 460 {
 461     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 462
 463     return mms->auto_kernel_cmdline;
 464 }
 465
 466 static void microvm_machine_set_auto_kernel_cmdline(Object *obj, bool value,
 467                                                     Error **errp)
 468 {
 469     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 470
 471     mms->auto_kernel_cmdline = value;
 472 }
 473
 474 static void microvm_machine_initfn(Object *obj)
 475 {
 476     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 477
 478     /* Configuration */
 479     mms->pic = ON_OFF_AUTO_AUTO;
 480     mms->pit = ON_OFF_AUTO_AUTO;
 481     mms->rtc = ON_OFF_AUTO_AUTO;
 482     mms->isa_serial = true;
 483     mms->option_roms = true;
 484     mms->auto_kernel_cmdline = true;
 485
 486     /* State */
 487     mms->kernel_cmdline_fixed = false;
 488 }
 489
 490 static void microvm_class_init(ObjectClass *oc, void *data)
 491 {
 492     MachineClass *mc = MACHINE_CLASS(oc);
 493
 494     mc->init = microvm_machine_state_init;
 495
 496     mc->family = "microvm_i386";
 497     mc->desc = "microvm (i386)";
 498     mc->units_per_default_bus = 1;
 499     mc->no_floppy = 1;
 500     mc->max_cpus = 288;
 501     mc->has_hotpluggable_cpus = false;
 502     mc->auto_enable_numa_with_memhp = false;
 503     mc->default_cpu_type = TARGET_DEFAULT_CPU_TYPE;
 504     mc->nvdimm_supported = false;
 505
 506     /* Avoid relying too much on kernel components */
 507     mc->default_kernel_irqchip_split = true;
 508
 509     /* Machine class handlers */
 510     mc->reset = microvm_machine_reset;
 511
 512     object_class_property_add(oc, MICROVM_MACHINE_PIC, "OnOffAuto",
 513                               microvm_machine_get_pic,
 514                               microvm_machine_set_pic,
 515                               NULL, NULL, &error_abort);
 516     object_class_property_set_description(oc, MICROVM_MACHINE_PIC,
 517         "Enable i8259 PIC", &error_abort);
 518
 519     object_class_property_add(oc, MICROVM_MACHINE_PIT, "OnOffAuto",
 520                               microvm_machine_get_pit,
 521                               microvm_machine_set_pit,
 522                               NULL, NULL, &error_abort);
 523     object_class_property_set_description(oc, MICROVM_MACHINE_PIT,
 524         "Enable i8254 PIT", &error_abort);
 525
 526     object_class_property_add(oc, MICROVM_MACHINE_RTC, "OnOffAuto",
 527                               microvm_machine_get_rtc,
 528                               microvm_machine_set_rtc,
 529                               NULL, NULL, &error_abort);
 530     object_class_property_set_description(oc, MICROVM_MACHINE_RTC,
 531         "Enable MC146818 RTC", &error_abort);
 532
 533     object_class_property_add_bool(oc, MICROVM_MACHINE_ISA_SERIAL,
 534                                    microvm_machine_get_isa_serial,
 535                                    microvm_machine_set_isa_serial,
 536                                    &error_abort);
 537     object_class_property_set_description(oc, MICROVM_MACHINE_ISA_SERIAL,
 538         "Set off to disable the instantiation an ISA serial port",
 539         &error_abort);
 540
 541     object_class_property_add_bool(oc, MICROVM_MACHINE_OPTION_ROMS,
 542                                    microvm_machine_get_option_roms,
 543                                    microvm_machine_set_option_roms,
 544                                    &error_abort);
 545     object_class_property_set_description(oc, MICROVM_MACHINE_OPTION_ROMS,
 546         "Set off to disable loading option ROMs", &error_abort);
 547
 548     object_class_property_add_bool(oc, MICROVM_MACHINE_AUTO_KERNEL_CMDLINE,
 549                                    microvm_machine_get_auto_kernel_cmdline,
 550                                    microvm_machine_set_auto_kernel_cmdline,
 551                                    &error_abort);
 552     object_class_property_set_description(oc,
 553         MICROVM_MACHINE_AUTO_KERNEL_CMDLINE,
 554         "Set off to disable adding virtio-mmio devices to the kernel cmdline",
 555         &error_abort);
 556 }
 557
 558 static const TypeInfo microvm_machine_info = {
 559     .name          = TYPE_MICROVM_MACHINE,
 560     .parent        = TYPE_X86_MACHINE,
 561     .instance_size = sizeof(MicrovmMachineState),
 562     .instance_init = microvm_machine_initfn,
 563     .class_size    = sizeof(MicrovmMachineClass),
 564     .class_init    = microvm_class_init,
 565     .interfaces = (InterfaceInfo[]) {
 566          { }
 567     },
 568 };
 569
 570 static void microvm_machine_init(void)
 571 {
 572     type_register_static(&microvm_machine_info);
 573 }
 574 type_init(microvm_machine_init);