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_below_4g, *ram_above_4g;
 171     MemoryRegion *system_memory = get_system_memory();
 172     FWCfgState *fw_cfg;
 173     ram_addr_t lowmem = 0xc0000000; /* 3G */
 174     int i;
 175
 176     if (machine->ram_size > lowmem) {
 177         x86ms->above_4g_mem_size = machine->ram_size - lowmem;
 178         x86ms->below_4g_mem_size = lowmem;
 179     } else {
 180         x86ms->above_4g_mem_size = 0;
 181         x86ms->below_4g_mem_size = machine->ram_size;
 182     }
 183
 184     ram_below_4g = g_malloc(sizeof(*ram_below_4g));
 185     memory_region_init_alias(ram_below_4g, NULL, "ram-below-4g", machine->ram,
 186                              0, x86ms->below_4g_mem_size);
 187     memory_region_add_subregion(system_memory, 0, ram_below_4g);
 188
 189     e820_add_entry(0, x86ms->below_4g_mem_size, E820_RAM);
 190
 191     if (x86ms->above_4g_mem_size > 0) {
 192         ram_above_4g = g_malloc(sizeof(*ram_above_4g));
 193         memory_region_init_alias(ram_above_4g, NULL, "ram-above-4g",
 194                                  machine->ram,
 195                                  x86ms->below_4g_mem_size,
 196                                  x86ms->above_4g_mem_size);
 197         memory_region_add_subregion(system_memory, 0x100000000ULL,
 198                                     ram_above_4g);
 199         e820_add_entry(0x100000000ULL, x86ms->above_4g_mem_size, E820_RAM);
 200     }
 201
 202     fw_cfg = fw_cfg_init_io_dma(FW_CFG_IO_BASE, FW_CFG_IO_BASE + 4,
 203                                 &address_space_memory);
 204
 205     fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, machine->smp.cpus);
 206     fw_cfg_add_i16(fw_cfg, FW_CFG_MAX_CPUS, machine->smp.max_cpus);
 207     fw_cfg_add_i64(fw_cfg, FW_CFG_RAM_SIZE, (uint64_t)machine->ram_size);
 208     fw_cfg_add_i32(fw_cfg, FW_CFG_IRQ0_OVERRIDE, kvm_allows_irq0_override());
 209     fw_cfg_add_bytes(fw_cfg, FW_CFG_E820_TABLE,
 210                      &e820_reserve, sizeof(e820_reserve));
 211     fw_cfg_add_file(fw_cfg, "etc/e820", e820_table,
 212                     sizeof(struct e820_entry) * e820_get_num_entries());
 213
 214     rom_set_fw(fw_cfg);
 215
 216     if (machine->kernel_filename != NULL) {
 217         x86_load_linux(x86ms, fw_cfg, 0, true, true);
 218     }
 219
 220     if (mms->option_roms) {
 221         for (i = 0; i < nb_option_roms; i++) {
 222             rom_add_option(option_rom[i].name, option_rom[i].bootindex);
 223         }
 224     }
 225
 226     x86ms->fw_cfg = fw_cfg;
 227     x86ms->ioapic_as = &address_space_memory;
 228 }
 229
 230 static gchar *microvm_get_mmio_cmdline(gchar *name)
 231 {
 232     gchar *cmdline;
 233     gchar *separator;
 234     long int index;
 235     int ret;
 236
 237     separator = g_strrstr(name, ".");
 238     if (!separator) {
 239         return NULL;
 240     }
 241
 242     if (qemu_strtol(separator + 1, NULL, 10, &index) != 0) {
 243         return NULL;
 244     }
 245
 246     cmdline = g_malloc0(VIRTIO_CMDLINE_MAXLEN);
 247     ret = g_snprintf(cmdline, VIRTIO_CMDLINE_MAXLEN,
 248                      " virtio_mmio.device=512@0x%lx:%ld",
 249                      VIRTIO_MMIO_BASE + index * 512,
 250                      VIRTIO_IRQ_BASE + index);
 251     if (ret < 0 || ret >= VIRTIO_CMDLINE_MAXLEN) {
 252         g_free(cmdline);
 253         return NULL;
 254     }
 255
 256     return cmdline;
 257 }
 258
 259 static void microvm_fix_kernel_cmdline(MachineState *machine)
 260 {
 261     X86MachineState *x86ms = X86_MACHINE(machine);
 262     BusState *bus;
 263     BusChild *kid;
 264     char *cmdline;
 265
 266     /*
 267      * Find MMIO transports with attached devices, and add them to the kernel
 268      * command line.
 269      *
 270      * Yes, this is a hack, but one that heavily improves the UX without
 271      * introducing any significant issues.
 272      */
 273     cmdline = g_strdup(machine->kernel_cmdline);
 274     bus = sysbus_get_default();
 275     QTAILQ_FOREACH(kid, &bus->children, sibling) {
 276         DeviceState *dev = kid->child;
 277         ObjectClass *class = object_get_class(OBJECT(dev));
 278
 279         if (class == object_class_by_name(TYPE_VIRTIO_MMIO)) {
 280             VirtIOMMIOProxy *mmio = VIRTIO_MMIO(OBJECT(dev));
 281             VirtioBusState *mmio_virtio_bus = &mmio->bus;
 282             BusState *mmio_bus = &mmio_virtio_bus->parent_obj;
 283
 284             if (!QTAILQ_EMPTY(&mmio_bus->children)) {
 285                 gchar *mmio_cmdline = microvm_get_mmio_cmdline(mmio_bus->name);
 286                 if (mmio_cmdline) {
 287                     char *newcmd = g_strjoin(NULL, cmdline, mmio_cmdline, NULL);
 288                     g_free(mmio_cmdline);
 289                     g_free(cmdline);
 290                     cmdline = newcmd;
 291                 }
 292             }
 293         }
 294     }
 295
 296     fw_cfg_modify_i32(x86ms->fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(cmdline) + 1);
 297     fw_cfg_modify_string(x86ms->fw_cfg, FW_CFG_CMDLINE_DATA, cmdline);
 298
 299     g_free(cmdline);
 300 }
 301
 302 static void microvm_machine_state_init(MachineState *machine)
 303 {
 304     MicrovmMachineState *mms = MICROVM_MACHINE(machine);
 305     X86MachineState *x86ms = X86_MACHINE(machine);
 306     Error *local_err = NULL;
 307
 308     microvm_memory_init(mms);
 309
 310     x86_cpus_init(x86ms, CPU_VERSION_LATEST);
 311     if (local_err) {
 312         error_report_err(local_err);
 313         exit(1);
 314     }
 315
 316     microvm_devices_init(mms);
 317 }
 318
 319 static void microvm_machine_reset(MachineState *machine)
 320 {
 321     MicrovmMachineState *mms = MICROVM_MACHINE(machine);
 322     CPUState *cs;
 323     X86CPU *cpu;
 324
 325     if (machine->kernel_filename != NULL &&
 326         mms->auto_kernel_cmdline && !mms->kernel_cmdline_fixed) {
 327         microvm_fix_kernel_cmdline(machine);
 328         mms->kernel_cmdline_fixed = true;
 329     }
 330
 331     qemu_devices_reset();
 332
 333     CPU_FOREACH(cs) {
 334         cpu = X86_CPU(cs);
 335
 336         if (cpu->apic_state) {
 337             device_legacy_reset(cpu->apic_state);
 338         }
 339     }
 340 }
 341
 342 static void microvm_machine_get_pic(Object *obj, Visitor *v, const char *name,
 343                                     void *opaque, Error **errp)
 344 {
 345     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 346     OnOffAuto pic = mms->pic;
 347
 348     visit_type_OnOffAuto(v, name, &pic, errp);
 349 }
 350
 351 static void microvm_machine_set_pic(Object *obj, Visitor *v, const char *name,
 352                                     void *opaque, Error **errp)
 353 {
 354     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 355
 356     visit_type_OnOffAuto(v, name, &mms->pic, errp);
 357 }
 358
 359 static void microvm_machine_get_pit(Object *obj, Visitor *v, const char *name,
 360                                     void *opaque, Error **errp)
 361 {
 362     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 363     OnOffAuto pit = mms->pit;
 364
 365     visit_type_OnOffAuto(v, name, &pit, errp);
 366 }
 367
 368 static void microvm_machine_set_pit(Object *obj, Visitor *v, const char *name,
 369                                     void *opaque, Error **errp)
 370 {
 371     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 372
 373     visit_type_OnOffAuto(v, name, &mms->pit, errp);
 374 }
 375
 376 static void microvm_machine_get_rtc(Object *obj, Visitor *v, const char *name,
 377                                     void *opaque, Error **errp)
 378 {
 379     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 380     OnOffAuto rtc = mms->rtc;
 381
 382     visit_type_OnOffAuto(v, name, &rtc, errp);
 383 }
 384
 385 static void microvm_machine_set_rtc(Object *obj, Visitor *v, const char *name,
 386                                     void *opaque, Error **errp)
 387 {
 388     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 389
 390     visit_type_OnOffAuto(v, name, &mms->rtc, errp);
 391 }
 392
 393 static bool microvm_machine_get_isa_serial(Object *obj, Error **errp)
 394 {
 395     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 396
 397     return mms->isa_serial;
 398 }
 399
 400 static void microvm_machine_set_isa_serial(Object *obj, bool value,
 401                                            Error **errp)
 402 {
 403     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 404
 405     mms->isa_serial = value;
 406 }
 407
 408 static bool microvm_machine_get_option_roms(Object *obj, Error **errp)
 409 {
 410     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 411
 412     return mms->option_roms;
 413 }
 414
 415 static void microvm_machine_set_option_roms(Object *obj, bool value,
 416                                             Error **errp)
 417 {
 418     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 419
 420     mms->option_roms = value;
 421 }
 422
 423 static bool microvm_machine_get_auto_kernel_cmdline(Object *obj, Error **errp)
 424 {
 425     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 426
 427     return mms->auto_kernel_cmdline;
 428 }
 429
 430 static void microvm_machine_set_auto_kernel_cmdline(Object *obj, bool value,
 431                                                     Error **errp)
 432 {
 433     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 434
 435     mms->auto_kernel_cmdline = value;
 436 }
 437
 438 static void microvm_machine_initfn(Object *obj)
 439 {
 440     MicrovmMachineState *mms = MICROVM_MACHINE(obj);
 441
 442     /* Configuration */
 443     mms->pic = ON_OFF_AUTO_AUTO;
 444     mms->pit = ON_OFF_AUTO_AUTO;
 445     mms->rtc = ON_OFF_AUTO_AUTO;
 446     mms->isa_serial = true;
 447     mms->option_roms = true;
 448     mms->auto_kernel_cmdline = true;
 449
 450     /* State */
 451     mms->kernel_cmdline_fixed = false;
 452 }
 453
 454 static void microvm_class_init(ObjectClass *oc, void *data)
 455 {
 456     MachineClass *mc = MACHINE_CLASS(oc);
 457
 458     mc->init = microvm_machine_state_init;
 459
 460     mc->family = "microvm_i386";
 461     mc->desc = "microvm (i386)";
 462     mc->units_per_default_bus = 1;
 463     mc->no_floppy = 1;
 464     mc->max_cpus = 288;
 465     mc->has_hotpluggable_cpus = false;
 466     mc->auto_enable_numa_with_memhp = false;
 467     mc->auto_enable_numa_with_memdev = false;
 468     mc->default_cpu_type = TARGET_DEFAULT_CPU_TYPE;
 469     mc->nvdimm_supported = false;
 470     mc->default_ram_id = "microvm.ram";
 471
 472     /* Avoid relying too much on kernel components */
 473     mc->default_kernel_irqchip_split = true;
 474
 475     /* Machine class handlers */
 476     mc->reset = microvm_machine_reset;
 477
 478     object_class_property_add(oc, MICROVM_MACHINE_PIC, "OnOffAuto",
 479                               microvm_machine_get_pic,
 480                               microvm_machine_set_pic,
 481                               NULL, NULL);
 482     object_class_property_set_description(oc, MICROVM_MACHINE_PIC,
 483         "Enable i8259 PIC");
 484
 485     object_class_property_add(oc, MICROVM_MACHINE_PIT, "OnOffAuto",
 486                               microvm_machine_get_pit,
 487                               microvm_machine_set_pit,
 488                               NULL, NULL);
 489     object_class_property_set_description(oc, MICROVM_MACHINE_PIT,
 490         "Enable i8254 PIT");
 491
 492     object_class_property_add(oc, MICROVM_MACHINE_RTC, "OnOffAuto",
 493                               microvm_machine_get_rtc,
 494                               microvm_machine_set_rtc,
 495                               NULL, NULL);
 496     object_class_property_set_description(oc, MICROVM_MACHINE_RTC,
 497         "Enable MC146818 RTC");
 498
 499     object_class_property_add_bool(oc, MICROVM_MACHINE_ISA_SERIAL,
 500                                    microvm_machine_get_isa_serial,
 501                                    microvm_machine_set_isa_serial);
 502     object_class_property_set_description(oc, MICROVM_MACHINE_ISA_SERIAL,
 503         "Set off to disable the instantiation an ISA serial port");
 504
 505     object_class_property_add_bool(oc, MICROVM_MACHINE_OPTION_ROMS,
 506                                    microvm_machine_get_option_roms,
 507                                    microvm_machine_set_option_roms);
 508     object_class_property_set_description(oc, MICROVM_MACHINE_OPTION_ROMS,
 509         "Set off to disable loading option ROMs");
 510
 511     object_class_property_add_bool(oc, MICROVM_MACHINE_AUTO_KERNEL_CMDLINE,
 512                                    microvm_machine_get_auto_kernel_cmdline,
 513                                    microvm_machine_set_auto_kernel_cmdline);
 514     object_class_property_set_description(oc,
 515         MICROVM_MACHINE_AUTO_KERNEL_CMDLINE,
 516         "Set off to disable adding virtio-mmio devices to the kernel cmdline");
 517 }
 518
 519 static const TypeInfo microvm_machine_info = {
 520     .name          = TYPE_MICROVM_MACHINE,
 521     .parent        = TYPE_X86_MACHINE,
 522     .instance_size = sizeof(MicrovmMachineState),
 523     .instance_init = microvm_machine_initfn,
 524     .class_size    = sizeof(MicrovmMachineClass),
 525     .class_init    = microvm_class_init,
 526     .interfaces = (InterfaceInfo[]) {
 527          { }
 528     },
 529 };
 530
 531 static void microvm_machine_init(void)
 532 {
 533     type_register_static(&microvm_machine_info);
 534 }
 535 type_init(microvm_machine_init);