hw/pc.c

   1 /*
   2  * QEMU PC System Emulator
   3  *
   4  * Copyright (c) 2003-2004 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "hw.h"
  25 #include "pc.h"
  26 #include "apic.h"
  27 #include "fdc.h"
  28 #include "ide.h"
  29 #include "pci.h"
  30 #include "vmware_vga.h"
  31 #include "monitor.h"
  32 #include "fw_cfg.h"
  33 #include "hpet_emul.h"
  34 #include "smbios.h"
  35 #include "loader.h"
  36 #include "elf.h"
  37 #include "multiboot.h"
  38 #include "mc146818rtc.h"
  39 #include "i8254.h"
  40 #include "pcspk.h"
  41 #include "msi.h"
  42 #include "sysbus.h"
  43 #include "sysemu.h"
  44 #include "kvm.h"
  45 #include "xen.h"
  46 #include "blockdev.h"
  47 #include "ui/qemu-spice.h"
  48 #include "memory.h"
  49 #include "exec-memory.h"
  50
  51 /* output Bochs bios info messages */
  52 //#define DEBUG_BIOS
  53
  54 /* debug PC/ISA interrupts */
  55 //#define DEBUG_IRQ
  56
  57 #ifdef DEBUG_IRQ
  58 #define DPRINTF(fmt, ...)                                       \
  59     do { printf("CPUIRQ: " fmt , ## __VA_ARGS__); } while (0)
  60 #else
  61 #define DPRINTF(fmt, ...)
  62 #endif
  63
  64 /* Leave a chunk of memory at the top of RAM for the BIOS ACPI tables.  */
  65 #define ACPI_DATA_SIZE       0x10000
  66 #define BIOS_CFG_IOPORT 0x510
  67 #define FW_CFG_ACPI_TABLES (FW_CFG_ARCH_LOCAL + 0)
  68 #define FW_CFG_SMBIOS_ENTRIES (FW_CFG_ARCH_LOCAL + 1)
  69 #define FW_CFG_IRQ0_OVERRIDE (FW_CFG_ARCH_LOCAL + 2)
  70 #define FW_CFG_E820_TABLE (FW_CFG_ARCH_LOCAL + 3)
  71 #define FW_CFG_HPET (FW_CFG_ARCH_LOCAL + 4)
  72
  73 #define MSI_ADDR_BASE 0xfee00000
  74
  75 #define E820_NR_ENTRIES         16
  76
  77 struct e820_entry {
  78     uint64_t address;
  79     uint64_t length;
  80     uint32_t type;
  81 } QEMU_PACKED __attribute((__aligned__(4)));
  82
  83 struct e820_table {
  84     uint32_t count;
  85     struct e820_entry entry[E820_NR_ENTRIES];
  86 } QEMU_PACKED __attribute((__aligned__(4)));
  87
  88 static struct e820_table e820_table;
  89 struct hpet_fw_config hpet_cfg = {.count = UINT8_MAX};
  90
  91 void gsi_handler(void *opaque, int n, int level)
  92 {
  93     GSIState *s = opaque;
  94
  95     DPRINTF("pc: %s GSI %d\n", level ? "raising" : "lowering", n);
  96     if (n < ISA_NUM_IRQS) {
  97         qemu_set_irq(s->i8259_irq[n], level);
  98     }
  99     qemu_set_irq(s->ioapic_irq[n], level);
 100 }
 101
 102 static void ioport80_write(void *opaque, uint32_t addr, uint32_t data)
 103 {
 104 }
 105
 106 /* MSDOS compatibility mode FPU exception support */
 107 static qemu_irq ferr_irq;
 108
 109 void pc_register_ferr_irq(qemu_irq irq)
 110 {
 111     ferr_irq = irq;
 112 }
 113
 114 /* XXX: add IGNNE support */
 115 void cpu_set_ferr(CPUX86State *s)
 116 {
 117     qemu_irq_raise(ferr_irq);
 118 }
 119
 120 static void ioportF0_write(void *opaque, uint32_t addr, uint32_t data)
 121 {
 122     qemu_irq_lower(ferr_irq);
 123 }
 124
 125 /* TSC handling */
 126 uint64_t cpu_get_tsc(CPUX86State *env)
 127 {
 128     return cpu_get_ticks();
 129 }
 130
 131 /* SMM support */
 132
 133 static cpu_set_smm_t smm_set;
 134 static void *smm_arg;
 135
 136 void cpu_smm_register(cpu_set_smm_t callback, void *arg)
 137 {
 138     assert(smm_set == NULL);
 139     assert(smm_arg == NULL);
 140     smm_set = callback;
 141     smm_arg = arg;
 142 }
 143
 144 void cpu_smm_update(CPUX86State *env)
 145 {
 146     if (smm_set && smm_arg && env == first_cpu)
 147         smm_set(!!(env->hflags & HF_SMM_MASK), smm_arg);
 148 }
 149
 150
 151 /* IRQ handling */
 152 int cpu_get_pic_interrupt(CPUX86State *env)
 153 {
 154     int intno;
 155
 156     intno = apic_get_interrupt(env->apic_state);
 157     if (intno >= 0) {
 158         return intno;
 159     }
 160     /* read the irq from the PIC */
 161     if (!apic_accept_pic_intr(env->apic_state)) {
 162         return -1;
 163     }
 164
 165     intno = pic_read_irq(isa_pic);
 166     return intno;
 167 }
 168
 169 static void pic_irq_request(void *opaque, int irq, int level)
 170 {
 171     CPUX86State *env = first_cpu;
 172
 173     DPRINTF("pic_irqs: %s irq %d\n", level? "raise" : "lower", irq);
 174     if (env->apic_state) {
 175         while (env) {
 176             if (apic_accept_pic_intr(env->apic_state)) {
 177                 apic_deliver_pic_intr(env->apic_state, level);
 178             }
 179             env = env->next_cpu;
 180         }
 181     } else {
 182         if (level)
 183             cpu_interrupt(env, CPU_INTERRUPT_HARD);
 184         else
 185             cpu_reset_interrupt(env, CPU_INTERRUPT_HARD);
 186     }
 187 }
 188
 189 /* PC cmos mappings */
 190
 191 #define REG_EQUIPMENT_BYTE          0x14
 192
 193 static int cmos_get_fd_drive_type(FDriveType fd0)
 194 {
 195     int val;
 196
 197     switch (fd0) {
 198     case FDRIVE_DRV_144:
 199         /* 1.44 Mb 3"5 drive */
 200         val = 4;
 201         break;
 202     case FDRIVE_DRV_288:
 203         /* 2.88 Mb 3"5 drive */
 204         val = 5;
 205         break;
 206     case FDRIVE_DRV_120:
 207         /* 1.2 Mb 5"5 drive */
 208         val = 2;
 209         break;
 210     case FDRIVE_DRV_NONE:
 211     default:
 212         val = 0;
 213         break;
 214     }
 215     return val;
 216 }
 217
 218 static void cmos_init_hd(int type_ofs, int info_ofs, BlockDriverState *hd,
 219                          ISADevice *s)
 220 {
 221     int cylinders, heads, sectors;
 222     bdrv_get_geometry_hint(hd, &cylinders, &heads, &sectors);
 223     rtc_set_memory(s, type_ofs, 47);
 224     rtc_set_memory(s, info_ofs, cylinders);
 225     rtc_set_memory(s, info_ofs + 1, cylinders >> 8);
 226     rtc_set_memory(s, info_ofs + 2, heads);
 227     rtc_set_memory(s, info_ofs + 3, 0xff);
 228     rtc_set_memory(s, info_ofs + 4, 0xff);
 229     rtc_set_memory(s, info_ofs + 5, 0xc0 | ((heads > 8) << 3));
 230     rtc_set_memory(s, info_ofs + 6, cylinders);
 231     rtc_set_memory(s, info_ofs + 7, cylinders >> 8);
 232     rtc_set_memory(s, info_ofs + 8, sectors);
 233 }
 234
 235 /* convert boot_device letter to something recognizable by the bios */
 236 static int boot_device2nibble(char boot_device)
 237 {
 238     switch(boot_device) {
 239     case 'a':
 240     case 'b':
 241         return 0x01; /* floppy boot */
 242     case 'c':
 243         return 0x02; /* hard drive boot */
 244     case 'd':
 245         return 0x03; /* CD-ROM boot */
 246     case 'n':
 247         return 0x04; /* Network boot */
 248     }
 249     return 0;
 250 }
 251
 252 static int set_boot_dev(ISADevice *s, const char *boot_device, int fd_bootchk)
 253 {
 254 #define PC_MAX_BOOT_DEVICES 3
 255     int nbds, bds[3] = { 0, };
 256     int i;
 257
 258     nbds = strlen(boot_device);
 259     if (nbds > PC_MAX_BOOT_DEVICES) {
 260         error_report("Too many boot devices for PC");
 261         return(1);
 262     }
 263     for (i = 0; i < nbds; i++) {
 264         bds[i] = boot_device2nibble(boot_device[i]);
 265         if (bds[i] == 0) {
 266             error_report("Invalid boot device for PC: '%c'",
 267                          boot_device[i]);
 268             return(1);
 269         }
 270     }
 271     rtc_set_memory(s, 0x3d, (bds[1] << 4) | bds[0]);
 272     rtc_set_memory(s, 0x38, (bds[2] << 4) | (fd_bootchk ? 0x0 : 0x1));
 273     return(0);
 274 }
 275
 276 static int pc_boot_set(void *opaque, const char *boot_device)
 277 {
 278     return set_boot_dev(opaque, boot_device, 0);
 279 }
 280
 281 typedef struct pc_cmos_init_late_arg {
 282     ISADevice *rtc_state;
 283     BusState *idebus0, *idebus1;
 284 } pc_cmos_init_late_arg;
 285
 286 static void pc_cmos_init_late(void *opaque)
 287 {
 288     pc_cmos_init_late_arg *arg = opaque;
 289     ISADevice *s = arg->rtc_state;
 290     int val;
 291     BlockDriverState *hd_table[4];
 292     int i;
 293
 294     ide_get_bs(hd_table, arg->idebus0);
 295     ide_get_bs(hd_table + 2, arg->idebus1);
 296
 297     rtc_set_memory(s, 0x12, (hd_table[0] ? 0xf0 : 0) | (hd_table[1] ? 0x0f : 0));
 298     if (hd_table[0])
 299         cmos_init_hd(0x19, 0x1b, hd_table[0], s);
 300     if (hd_table[1])
 301         cmos_init_hd(0x1a, 0x24, hd_table[1], s);
 302
 303     val = 0;
 304     for (i = 0; i < 4; i++) {
 305         if (hd_table[i]) {
 306             int cylinders, heads, sectors, translation;
 307             /* NOTE: bdrv_get_geometry_hint() returns the physical
 308                 geometry.  It is always such that: 1 <= sects <= 63, 1
 309                 <= heads <= 16, 1 <= cylinders <= 16383. The BIOS
 310                 geometry can be different if a translation is done. */
 311             translation = bdrv_get_translation_hint(hd_table[i]);
 312             if (translation == BIOS_ATA_TRANSLATION_AUTO) {
 313                 bdrv_get_geometry_hint(hd_table[i], &cylinders, &heads, &sectors);
 314                 if (cylinders <= 1024 && heads <= 16 && sectors <= 63) {
 315                     /* No translation. */
 316                     translation = 0;
 317                 } else {
 318                     /* LBA translation. */
 319                     translation = 1;
 320                 }
 321             } else {
 322                 translation--;
 323             }
 324             val |= translation << (i * 2);
 325         }
 326     }
 327     rtc_set_memory(s, 0x39, val);
 328
 329     qemu_unregister_reset(pc_cmos_init_late, opaque);
 330 }
 331
 332 void pc_cmos_init(ram_addr_t ram_size, ram_addr_t above_4g_mem_size,
 333                   const char *boot_device,
 334                   ISADevice *floppy, BusState *idebus0, BusState *idebus1,
 335                   ISADevice *s)
 336 {
 337     int val, nb, nb_heads, max_track, last_sect, i;
 338     FDriveType fd_type[2] = { FDRIVE_DRV_NONE, FDRIVE_DRV_NONE };
 339     FDriveRate rate;
 340     BlockDriverState *fd[MAX_FD];
 341     static pc_cmos_init_late_arg arg;
 342
 343     /* various important CMOS locations needed by PC/Bochs bios */
 344
 345     /* memory size */
 346     val = 640; /* base memory in K */
 347     rtc_set_memory(s, 0x15, val);
 348     rtc_set_memory(s, 0x16, val >> 8);
 349
 350     val = (ram_size / 1024) - 1024;
 351     if (val > 65535)
 352         val = 65535;
 353     rtc_set_memory(s, 0x17, val);
 354     rtc_set_memory(s, 0x18, val >> 8);
 355     rtc_set_memory(s, 0x30, val);
 356     rtc_set_memory(s, 0x31, val >> 8);
 357
 358     if (above_4g_mem_size) {
 359         rtc_set_memory(s, 0x5b, (unsigned int)above_4g_mem_size >> 16);
 360         rtc_set_memory(s, 0x5c, (unsigned int)above_4g_mem_size >> 24);
 361         rtc_set_memory(s, 0x5d, (uint64_t)above_4g_mem_size >> 32);
 362     }
 363
 364     if (ram_size > (16 * 1024 * 1024))
 365         val = (ram_size / 65536) - ((16 * 1024 * 1024) / 65536);
 366     else
 367         val = 0;
 368     if (val > 65535)
 369         val = 65535;
 370     rtc_set_memory(s, 0x34, val);
 371     rtc_set_memory(s, 0x35, val >> 8);
 372
 373     /* set the number of CPU */
 374     rtc_set_memory(s, 0x5f, smp_cpus - 1);
 375
 376     /* set boot devices, and disable floppy signature check if requested */
 377     if (set_boot_dev(s, boot_device, fd_bootchk)) {
 378         exit(1);
 379     }
 380
 381     /* floppy type */
 382     if (floppy) {
 383         fdc_get_bs(fd, floppy);
 384         for (i = 0; i < 2; i++) {
 385             if (fd[i] && bdrv_is_inserted(fd[i])) {
 386                 bdrv_get_floppy_geometry_hint(fd[i], &nb_heads, &max_track,
 387                                               &last_sect, FDRIVE_DRV_NONE,
 388                                               &fd_type[i], &rate);
 389             }
 390         }
 391     }
 392     val = (cmos_get_fd_drive_type(fd_type[0]) << 4) |
 393         cmos_get_fd_drive_type(fd_type[1]);
 394     rtc_set_memory(s, 0x10, val);
 395
 396     val = 0;
 397     nb = 0;
 398     if (fd_type[0] < FDRIVE_DRV_NONE) {
 399         nb++;
 400     }
 401     if (fd_type[1] < FDRIVE_DRV_NONE) {
 402         nb++;
 403     }
 404     switch (nb) {
 405     case 0:
 406         break;
 407     case 1:
 408         val |= 0x01; /* 1 drive, ready for boot */
 409         break;
 410     case 2:
 411         val |= 0x41; /* 2 drives, ready for boot */
 412         break;
 413     }
 414     val |= 0x02; /* FPU is there */
 415     val |= 0x04; /* PS/2 mouse installed */
 416     rtc_set_memory(s, REG_EQUIPMENT_BYTE, val);
 417
 418     /* hard drives */
 419     arg.rtc_state = s;
 420     arg.idebus0 = idebus0;
 421     arg.idebus1 = idebus1;
 422     qemu_register_reset(pc_cmos_init_late, &arg);
 423 }
 424
 425 /* port 92 stuff: could be split off */
 426 typedef struct Port92State {
 427     ISADevice dev;
 428     MemoryRegion io;
 429     uint8_t outport;
 430     qemu_irq *a20_out;
 431 } Port92State;
 432
 433 static void port92_write(void *opaque, uint32_t addr, uint32_t val)
 434 {
 435     Port92State *s = opaque;
 436
 437     DPRINTF("port92: write 0x%02x\n", val);
 438     s->outport = val;
 439     qemu_set_irq(*s->a20_out, (val >> 1) & 1);
 440     if (val & 1) {
 441         qemu_system_reset_request();
 442     }
 443 }
 444
 445 static uint32_t port92_read(void *opaque, uint32_t addr)
 446 {
 447     Port92State *s = opaque;
 448     uint32_t ret;
 449
 450     ret = s->outport;
 451     DPRINTF("port92: read 0x%02x\n", ret);
 452     return ret;
 453 }
 454
 455 static void port92_init(ISADevice *dev, qemu_irq *a20_out)
 456 {
 457     Port92State *s = DO_UPCAST(Port92State, dev, dev);
 458
 459     s->a20_out = a20_out;
 460 }
 461
 462 static const VMStateDescription vmstate_port92_isa = {
 463     .name = "port92",
 464     .version_id = 1,
 465     .minimum_version_id = 1,
 466     .minimum_version_id_old = 1,
 467     .fields      = (VMStateField []) {
 468         VMSTATE_UINT8(outport, Port92State),
 469         VMSTATE_END_OF_LIST()
 470     }
 471 };
 472
 473 static void port92_reset(DeviceState *d)
 474 {
 475     Port92State *s = container_of(d, Port92State, dev.qdev);
 476
 477     s->outport &= ~1;
 478 }
 479
 480 static const MemoryRegionPortio port92_portio[] = {
 481     { 0, 1, 1, .read = port92_read, .write = port92_write },
 482     PORTIO_END_OF_LIST(),
 483 };
 484
 485 static const MemoryRegionOps port92_ops = {
 486     .old_portio = port92_portio
 487 };
 488
 489 static int port92_initfn(ISADevice *dev)
 490 {
 491     Port92State *s = DO_UPCAST(Port92State, dev, dev);
 492
 493     memory_region_init_io(&s->io, &port92_ops, s, "port92", 1);
 494     isa_register_ioport(dev, &s->io, 0x92);
 495
 496     s->outport = 0;
 497     return 0;
 498 }
 499
 500 static void port92_class_initfn(ObjectClass *klass, void *data)
 501 {
 502     DeviceClass *dc = DEVICE_CLASS(klass);
 503     ISADeviceClass *ic = ISA_DEVICE_CLASS(klass);
 504     ic->init = port92_initfn;
 505     dc->no_user = 1;
 506     dc->reset = port92_reset;
 507     dc->vmsd = &vmstate_port92_isa;
 508 }
 509
 510 static TypeInfo port92_info = {
 511     .name          = "port92",
 512     .parent        = TYPE_ISA_DEVICE,
 513     .instance_size = sizeof(Port92State),
 514     .class_init    = port92_class_initfn,
 515 };
 516
 517 static void port92_register_types(void)
 518 {
 519     type_register_static(&port92_info);
 520 }
 521
 522 type_init(port92_register_types)
 523
 524 static void handle_a20_line_change(void *opaque, int irq, int level)
 525 {
 526     CPUX86State *cpu = opaque;
 527
 528     /* XXX: send to all CPUs ? */
 529     /* XXX: add logic to handle multiple A20 line sources */
 530     cpu_x86_set_a20(cpu, level);
 531 }
 532
 533 /***********************************************************/
 534 /* Bochs BIOS debug ports */
 535
 536 static void bochs_bios_write(void *opaque, uint32_t addr, uint32_t val)
 537 {
 538     static const char shutdown_str[8] = "Shutdown";
 539     static int shutdown_index = 0;
 540
 541     switch(addr) {
 542         /* Bochs BIOS messages */
 543     case 0x400:
 544     case 0x401:
 545         /* used to be panic, now unused */
 546         break;
 547     case 0x402:
 548     case 0x403:
 549 #ifdef DEBUG_BIOS
 550         fprintf(stderr, "%c", val);
 551 #endif
 552         break;
 553     case 0x8900:
 554         /* same as Bochs power off */
 555         if (val == shutdown_str[shutdown_index]) {
 556             shutdown_index++;
 557             if (shutdown_index == 8) {
 558                 shutdown_index = 0;
 559                 qemu_system_shutdown_request();
 560             }
 561         } else {
 562             shutdown_index = 0;
 563         }
 564         break;
 565
 566         /* LGPL'ed VGA BIOS messages */
 567     case 0x501:
 568     case 0x502:
 569         exit((val << 1) | 1);
 570     case 0x500:
 571     case 0x503:
 572 #ifdef DEBUG_BIOS
 573         fprintf(stderr, "%c", val);
 574 #endif
 575         break;
 576     }
 577 }
 578
 579 int e820_add_entry(uint64_t address, uint64_t length, uint32_t type)
 580 {
 581     int index = le32_to_cpu(e820_table.count);
 582     struct e820_entry *entry;
 583
 584     if (index >= E820_NR_ENTRIES)
 585         return -EBUSY;
 586     entry = &e820_table.entry[index++];
 587
 588     entry->address = cpu_to_le64(address);
 589     entry->length = cpu_to_le64(length);
 590     entry->type = cpu_to_le32(type);
 591
 592     e820_table.count = cpu_to_le32(index);
 593     return index;
 594 }
 595
 596 static void *bochs_bios_init(void)
 597 {
 598     void *fw_cfg;
 599     uint8_t *smbios_table;
 600     size_t smbios_len;
 601     uint64_t *numa_fw_cfg;
 602     int i, j;
 603
 604     register_ioport_write(0x400, 1, 2, bochs_bios_write, NULL);
 605     register_ioport_write(0x401, 1, 2, bochs_bios_write, NULL);
 606     register_ioport_write(0x402, 1, 1, bochs_bios_write, NULL);
 607     register_ioport_write(0x403, 1, 1, bochs_bios_write, NULL);
 608     register_ioport_write(0x8900, 1, 1, bochs_bios_write, NULL);
 609
 610     register_ioport_write(0x501, 1, 1, bochs_bios_write, NULL);
 611     register_ioport_write(0x501, 1, 2, bochs_bios_write, NULL);
 612     register_ioport_write(0x502, 1, 2, bochs_bios_write, NULL);
 613     register_ioport_write(0x500, 1, 1, bochs_bios_write, NULL);
 614     register_ioport_write(0x503, 1, 1, bochs_bios_write, NULL);
 615
 616     fw_cfg = fw_cfg_init(BIOS_CFG_IOPORT, BIOS_CFG_IOPORT + 1, 0, 0);
 617
 618     fw_cfg_add_i32(fw_cfg, FW_CFG_ID, 1);
 619     fw_cfg_add_i64(fw_cfg, FW_CFG_RAM_SIZE, (uint64_t)ram_size);
 620     fw_cfg_add_bytes(fw_cfg, FW_CFG_ACPI_TABLES, (uint8_t *)acpi_tables,
 621                      acpi_tables_len);
 622     fw_cfg_add_i32(fw_cfg, FW_CFG_IRQ0_OVERRIDE, kvm_allows_irq0_override());
 623
 624     smbios_table = smbios_get_table(&smbios_len);
 625     if (smbios_table)
 626         fw_cfg_add_bytes(fw_cfg, FW_CFG_SMBIOS_ENTRIES,
 627                          smbios_table, smbios_len);
 628     fw_cfg_add_bytes(fw_cfg, FW_CFG_E820_TABLE, (uint8_t *)&e820_table,
 629                      sizeof(struct e820_table));
 630
 631     fw_cfg_add_bytes(fw_cfg, FW_CFG_HPET, (uint8_t *)&hpet_cfg,
 632                      sizeof(struct hpet_fw_config));
 633     /* allocate memory for the NUMA channel: one (64bit) word for the number
 634      * of nodes, one word for each VCPU->node and one word for each node to
 635      * hold the amount of memory.
 636      */
 637     numa_fw_cfg = g_malloc0((1 + max_cpus + nb_numa_nodes) * 8);
 638     numa_fw_cfg[0] = cpu_to_le64(nb_numa_nodes);
 639     for (i = 0; i < max_cpus; i++) {
 640         for (j = 0; j < nb_numa_nodes; j++) {
 641             if (node_cpumask[j] & (1 << i)) {
 642                 numa_fw_cfg[i + 1] = cpu_to_le64(j);
 643                 break;
 644             }
 645         }
 646     }
 647     for (i = 0; i < nb_numa_nodes; i++) {
 648         numa_fw_cfg[max_cpus + 1 + i] = cpu_to_le64(node_mem[i]);
 649     }
 650     fw_cfg_add_bytes(fw_cfg, FW_CFG_NUMA, (uint8_t *)numa_fw_cfg,
 651                      (1 + max_cpus + nb_numa_nodes) * 8);
 652
 653     return fw_cfg;
 654 }
 655
 656 static long get_file_size(FILE *f)
 657 {
 658     long where, size;
 659
 660     /* XXX: on Unix systems, using fstat() probably makes more sense */
 661
 662     where = ftell(f);
 663     fseek(f, 0, SEEK_END);
 664     size = ftell(f);
 665     fseek(f, where, SEEK_SET);
 666
 667     return size;
 668 }
 669
 670 static void load_linux(void *fw_cfg,
 671                        const char *kernel_filename,
 672                        const char *initrd_filename,
 673                        const char *kernel_cmdline,
 674                        target_phys_addr_t max_ram_size)
 675 {
 676     uint16_t protocol;
 677     int setup_size, kernel_size, initrd_size = 0, cmdline_size;
 678     uint32_t initrd_max;
 679     uint8_t header[8192], *setup, *kernel, *initrd_data;
 680     target_phys_addr_t real_addr, prot_addr, cmdline_addr, initrd_addr = 0;
 681     FILE *f;
 682     char *vmode;
 683
 684     /* Align to 16 bytes as a paranoia measure */
 685     cmdline_size = (strlen(kernel_cmdline)+16) & ~15;
 686
 687     /* load the kernel header */
 688     f = fopen(kernel_filename, "rb");
 689     if (!f || !(kernel_size = get_file_size(f)) ||
 690         fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) !=
 691         MIN(ARRAY_SIZE(header), kernel_size)) {
 692         fprintf(stderr, "qemu: could not load kernel '%s': %s\n",
 693                 kernel_filename, strerror(errno));
 694         exit(1);
 695     }
 696
 697     /* kernel protocol version */
 698 #if 0
 699     fprintf(stderr, "header magic: %#x\n", ldl_p(header+0x202));
 700 #endif
 701     if (ldl_p(header+0x202) == 0x53726448)
 702         protocol = lduw_p(header+0x206);
 703     else {
 704         /* This looks like a multiboot kernel. If it is, let's stop
 705            treating it like a Linux kernel. */
 706         if (load_multiboot(fw_cfg, f, kernel_filename, initrd_filename,
 707                            kernel_cmdline, kernel_size, header))
 708             return;
 709         protocol = 0;
 710     }
 711
 712     if (protocol < 0x200 || !(header[0x211] & 0x01)) {
 713         /* Low kernel */
 714         real_addr    = 0x90000;
 715         cmdline_addr = 0x9a000 - cmdline_size;
 716         prot_addr    = 0x10000;
 717     } else if (protocol < 0x202) {
 718         /* High but ancient kernel */
 719         real_addr    = 0x90000;
 720         cmdline_addr = 0x9a000 - cmdline_size;
 721         prot_addr    = 0x100000;
 722     } else {
 723         /* High and recent kernel */
 724         real_addr    = 0x10000;
 725         cmdline_addr = 0x20000;
 726         prot_addr    = 0x100000;
 727     }
 728
 729 #if 0
 730     fprintf(stderr,
 731             "qemu: real_addr     = 0x" TARGET_FMT_plx "\n"
 732             "qemu: cmdline_addr  = 0x" TARGET_FMT_plx "\n"
 733             "qemu: prot_addr     = 0x" TARGET_FMT_plx "\n",
 734             real_addr,
 735             cmdline_addr,
 736             prot_addr);
 737 #endif
 738
 739     /* highest address for loading the initrd */
 740     if (protocol >= 0x203)
 741         initrd_max = ldl_p(header+0x22c);
 742     else
 743         initrd_max = 0x37ffffff;
 744
 745     if (initrd_max >= max_ram_size-ACPI_DATA_SIZE)
 746         initrd_max = max_ram_size-ACPI_DATA_SIZE-1;
 747
 748     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr);
 749     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(kernel_cmdline)+1);
 750     fw_cfg_add_bytes(fw_cfg, FW_CFG_CMDLINE_DATA,
 751                      (uint8_t*)strdup(kernel_cmdline),
 752                      strlen(kernel_cmdline)+1);
 753
 754     if (protocol >= 0x202) {
 755         stl_p(header+0x228, cmdline_addr);
 756     } else {
 757         stw_p(header+0x20, 0xA33F);
 758         stw_p(header+0x22, cmdline_addr-real_addr);
 759     }
 760
 761     /* handle vga= parameter */
 762     vmode = strstr(kernel_cmdline, "vga=");
 763     if (vmode) {
 764         unsigned int video_mode;
 765         /* skip "vga=" */
 766         vmode += 4;
 767         if (!strncmp(vmode, "normal", 6)) {
 768             video_mode = 0xffff;
 769         } else if (!strncmp(vmode, "ext", 3)) {
 770             video_mode = 0xfffe;
 771         } else if (!strncmp(vmode, "ask", 3)) {
 772             video_mode = 0xfffd;
 773         } else {
 774             video_mode = strtol(vmode, NULL, 0);
 775         }
 776         stw_p(header+0x1fa, video_mode);
 777     }
 778
 779     /* loader type */
 780     /* High nybble = B reserved for QEMU; low nybble is revision number.
 781        If this code is substantially changed, you may want to consider
 782        incrementing the revision. */
 783     if (protocol >= 0x200)
 784         header[0x210] = 0xB0;
 785
 786     /* heap */
 787     if (protocol >= 0x201) {
 788         header[0x211] |= 0x80;  /* CAN_USE_HEAP */
 789         stw_p(header+0x224, cmdline_addr-real_addr-0x200);
 790     }
 791
 792     /* load initrd */
 793     if (initrd_filename) {
 794         if (protocol < 0x200) {
 795             fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n");
 796             exit(1);
 797         }
 798
 799         initrd_size = get_image_size(initrd_filename);
 800         if (initrd_size < 0) {
 801             fprintf(stderr, "qemu: error reading initrd %s\n",
 802                     initrd_filename);
 803             exit(1);
 804         }
 805
 806         initrd_addr = (initrd_max-initrd_size) & ~4095;
 807
 808         initrd_data = g_malloc(initrd_size);
 809         load_image(initrd_filename, initrd_data);
 810
 811         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
 812         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
 813         fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size);
 814
 815         stl_p(header+0x218, initrd_addr);
 816         stl_p(header+0x21c, initrd_size);
 817     }
 818
 819     /* load kernel and setup */
 820     setup_size = header[0x1f1];
 821     if (setup_size == 0)
 822         setup_size = 4;
 823     setup_size = (setup_size+1)*512;
 824     kernel_size -= setup_size;
 825
 826     setup  = g_malloc(setup_size);
 827     kernel = g_malloc(kernel_size);
 828     fseek(f, 0, SEEK_SET);
 829     if (fread(setup, 1, setup_size, f) != setup_size) {
 830         fprintf(stderr, "fread() failed\n");
 831         exit(1);
 832     }
 833     if (fread(kernel, 1, kernel_size, f) != kernel_size) {
 834         fprintf(stderr, "fread() failed\n");
 835         exit(1);
 836     }
 837     fclose(f);
 838     memcpy(setup, header, MIN(sizeof(header), setup_size));
 839
 840     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr);
 841     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size);
 842     fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA, kernel, kernel_size);
 843
 844     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr);
 845     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size);
 846     fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size);
 847
 848     option_rom[nb_option_roms].name = "linuxboot.bin";
 849     option_rom[nb_option_roms].bootindex = 0;
 850     nb_option_roms++;
 851 }
 852
 853 #define NE2000_NB_MAX 6
 854
 855 static const int ne2000_io[NE2000_NB_MAX] = { 0x300, 0x320, 0x340, 0x360,
 856                                               0x280, 0x380 };
 857 static const int ne2000_irq[NE2000_NB_MAX] = { 9, 10, 11, 3, 4, 5 };
 858
 859 static const int parallel_io[MAX_PARALLEL_PORTS] = { 0x378, 0x278, 0x3bc };
 860 static const int parallel_irq[MAX_PARALLEL_PORTS] = { 7, 7, 7 };
 861
 862 void pc_init_ne2k_isa(ISABus *bus, NICInfo *nd)
 863 {
 864     static int nb_ne2k = 0;
 865
 866     if (nb_ne2k == NE2000_NB_MAX)
 867         return;
 868     isa_ne2000_init(bus, ne2000_io[nb_ne2k],
 869                     ne2000_irq[nb_ne2k], nd);
 870     nb_ne2k++;
 871 }
 872
 873 int cpu_is_bsp(CPUX86State *env)
 874 {
 875     /* We hard-wire the BSP to the first CPU. */
 876     return env->cpu_index == 0;
 877 }
 878
 879 DeviceState *cpu_get_current_apic(void)
 880 {
 881     if (cpu_single_env) {
 882         return cpu_single_env->apic_state;
 883     } else {
 884         return NULL;
 885     }
 886 }
 887
 888 static DeviceState *apic_init(void *env, uint8_t apic_id)
 889 {
 890     DeviceState *dev;
 891     static int apic_mapped;
 892
 893     if (kvm_irqchip_in_kernel()) {
 894         dev = qdev_create(NULL, "kvm-apic");
 895     } else if (xen_enabled()) {
 896         dev = qdev_create(NULL, "xen-apic");
 897     } else {
 898         dev = qdev_create(NULL, "apic");
 899     }
 900
 901     qdev_prop_set_uint8(dev, "id", apic_id);
 902     qdev_prop_set_ptr(dev, "cpu_env", env);
 903     qdev_init_nofail(dev);
 904
 905     /* XXX: mapping more APICs at the same memory location */
 906     if (apic_mapped == 0) {
 907         /* NOTE: the APIC is directly connected to the CPU - it is not
 908            on the global memory bus. */
 909         /* XXX: what if the base changes? */
 910         sysbus_mmio_map(sysbus_from_qdev(dev), 0, MSI_ADDR_BASE);
 911         apic_mapped = 1;
 912     }
 913
 914     /* KVM does not support MSI yet. */
 915     if (!kvm_irqchip_in_kernel()) {
 916         msi_supported = true;
 917     }
 918
 919     if (xen_msi_support()) {
 920         msi_supported = true;
 921     }
 922
 923     return dev;
 924 }
 925
 926 void pc_acpi_smi_interrupt(void *opaque, int irq, int level)
 927 {
 928     CPUX86State *s = opaque;
 929
 930     if (level) {
 931         cpu_interrupt(s, CPU_INTERRUPT_SMI);
 932     }
 933 }
 934
 935 static void pc_cpu_reset(void *opaque)
 936 {
 937     CPUX86State *env = opaque;
 938
 939     cpu_state_reset(env);
 940     env->halted = !cpu_is_bsp(env);
 941 }
 942
 943 static CPUX86State *pc_new_cpu(const char *cpu_model)
 944 {
 945     CPUX86State *env;
 946
 947     env = cpu_init(cpu_model);
 948     if (!env) {
 949         fprintf(stderr, "Unable to find x86 CPU definition\n");
 950         exit(1);
 951     }
 952     if ((env->cpuid_features & CPUID_APIC) || smp_cpus > 1) {
 953         env->apic_state = apic_init(env, env->cpuid_apic_id);
 954     }
 955     qemu_register_reset(pc_cpu_reset, env);
 956     pc_cpu_reset(env);
 957     return env;
 958 }
 959
 960 void pc_cpus_init(const char *cpu_model)
 961 {
 962     int i;
 963
 964     /* init CPUs */
 965     if (cpu_model == NULL) {
 966 #ifdef TARGET_X86_64
 967         cpu_model = "qemu64";
 968 #else
 969         cpu_model = "qemu32";
 970 #endif
 971     }
 972
 973     for(i = 0; i < smp_cpus; i++) {
 974         pc_new_cpu(cpu_model);
 975     }
 976 }
 977
 978 void pc_memory_init(MemoryRegion *system_memory,
 979                     const char *kernel_filename,
 980                     const char *kernel_cmdline,
 981                     const char *initrd_filename,
 982                     ram_addr_t below_4g_mem_size,
 983                     ram_addr_t above_4g_mem_size,
 984                     MemoryRegion *rom_memory,
 985                     MemoryRegion **ram_memory)
 986 {
 987     int linux_boot, i;
 988     MemoryRegion *ram, *option_rom_mr;
 989     MemoryRegion *ram_below_4g, *ram_above_4g;
 990     void *fw_cfg;
 991
 992     linux_boot = (kernel_filename != NULL);
 993
 994     /* Allocate RAM.  We allocate it as a single memory region and use
 995      * aliases to address portions of it, mostly for backwards compatibility
 996      * with older qemus that used qemu_ram_alloc().
 997      */
 998     ram = g_malloc(sizeof(*ram));
 999     memory_region_init_ram(ram, "pc.ram",
1000                            below_4g_mem_size + above_4g_mem_size);
1001     vmstate_register_ram_global(ram);
1002     *ram_memory = ram;
1003     ram_below_4g = g_malloc(sizeof(*ram_below_4g));
1004     memory_region_init_alias(ram_below_4g, "ram-below-4g", ram,
1005                              0, below_4g_mem_size);
1006     memory_region_add_subregion(system_memory, 0, ram_below_4g);
1007     if (above_4g_mem_size > 0) {
1008         ram_above_4g = g_malloc(sizeof(*ram_above_4g));
1009         memory_region_init_alias(ram_above_4g, "ram-above-4g", ram,
1010                                  below_4g_mem_size, above_4g_mem_size);
1011         memory_region_add_subregion(system_memory, 0x100000000ULL,
1012                                     ram_above_4g);
1013     }
1014
1015
1016     /* Initialize PC system firmware */
1017     pc_system_firmware_init(rom_memory);
1018
1019     option_rom_mr = g_malloc(sizeof(*option_rom_mr));
1020     memory_region_init_ram(option_rom_mr, "pc.rom", PC_ROM_SIZE);
1021     vmstate_register_ram_global(option_rom_mr);
1022     memory_region_add_subregion_overlap(rom_memory,
1023                                         PC_ROM_MIN_VGA,
1024                                         option_rom_mr,
1025                                         1);
1026
1027     fw_cfg = bochs_bios_init();
1028     rom_set_fw(fw_cfg);
1029
1030     if (linux_boot) {
1031         load_linux(fw_cfg, kernel_filename, initrd_filename, kernel_cmdline, below_4g_mem_size);
1032     }
1033
1034     for (i = 0; i < nb_option_roms; i++) {
1035         rom_add_option(option_rom[i].name, option_rom[i].bootindex);
1036     }
1037 }
1038
1039 qemu_irq *pc_allocate_cpu_irq(void)
1040 {
1041     return qemu_allocate_irqs(pic_irq_request, NULL, 1);
1042 }
1043
1044 DeviceState *pc_vga_init(ISABus *isa_bus, PCIBus *pci_bus)
1045 {
1046     DeviceState *dev = NULL;
1047
1048     if (cirrus_vga_enabled) {
1049         if (pci_bus) {
1050             dev = pci_cirrus_vga_init(pci_bus);
1051         } else {
1052             dev = &isa_create_simple(isa_bus, "isa-cirrus-vga")->qdev;
1053         }
1054     } else if (vmsvga_enabled) {
1055         if (pci_bus) {
1056             dev = pci_vmsvga_init(pci_bus);
1057         } else {
1058             fprintf(stderr, "%s: vmware_vga: no PCI bus\n", __FUNCTION__);
1059         }
1060 #ifdef CONFIG_SPICE
1061     } else if (qxl_enabled) {
1062         if (pci_bus) {
1063             dev = &pci_create_simple(pci_bus, -1, "qxl-vga")->qdev;
1064         } else {
1065             fprintf(stderr, "%s: qxl: no PCI bus\n", __FUNCTION__);
1066         }
1067 #endif
1068     } else if (std_vga_enabled) {
1069         if (pci_bus) {
1070             dev = pci_vga_init(pci_bus);
1071         } else {
1072             dev = isa_vga_init(isa_bus);
1073         }
1074     }
1075
1076     return dev;
1077 }
1078
1079 static void cpu_request_exit(void *opaque, int irq, int level)
1080 {
1081     CPUX86State *env = cpu_single_env;
1082
1083     if (env && level) {
1084         cpu_exit(env);
1085     }
1086 }
1087
1088 void pc_basic_device_init(ISABus *isa_bus, qemu_irq *gsi,
1089                           ISADevice **rtc_state,
1090                           ISADevice **floppy,
1091                           bool no_vmport)
1092 {
1093     int i;
1094     DriveInfo *fd[MAX_FD];
1095     DeviceState *hpet = NULL;
1096     int pit_isa_irq = 0;
1097     qemu_irq pit_alt_irq = NULL;
1098     qemu_irq rtc_irq = NULL;
1099     qemu_irq *a20_line;
1100     ISADevice *i8042, *port92, *vmmouse, *pit;
1101     qemu_irq *cpu_exit_irq;
1102
1103     register_ioport_write(0x80, 1, 1, ioport80_write, NULL);
1104
1105     register_ioport_write(0xf0, 1, 1, ioportF0_write, NULL);
1106
1107     /*
1108      * Check if an HPET shall be created.
1109      *
1110      * Without KVM_CAP_PIT_STATE2, we cannot switch off the in-kernel PIT
1111      * when the HPET wants to take over. Thus we have to disable the latter.
1112      */
1113     if (!no_hpet && (!kvm_irqchip_in_kernel() || kvm_has_pit_state2())) {
1114         hpet = sysbus_try_create_simple("hpet", HPET_BASE, NULL);
1115
1116         if (hpet) {
1117             for (i = 0; i < GSI_NUM_PINS; i++) {
1118                 sysbus_connect_irq(sysbus_from_qdev(hpet), i, gsi[i]);
1119             }
1120             pit_isa_irq = -1;
1121             pit_alt_irq = qdev_get_gpio_in(hpet, HPET_LEGACY_PIT_INT);
1122             rtc_irq = qdev_get_gpio_in(hpet, HPET_LEGACY_RTC_INT);
1123         }
1124     }
1125     *rtc_state = rtc_init(isa_bus, 2000, rtc_irq);
1126
1127     qemu_register_boot_set(pc_boot_set, *rtc_state);
1128
1129     if (kvm_irqchip_in_kernel()) {
1130         pit = kvm_pit_init(isa_bus, 0x40);
1131     } else {
1132         pit = pit_init(isa_bus, 0x40, pit_isa_irq, pit_alt_irq);
1133     }
1134     if (hpet) {
1135         /* connect PIT to output control line of the HPET */
1136         qdev_connect_gpio_out(hpet, 0, qdev_get_gpio_in(&pit->qdev, 0));
1137     }
1138     pcspk_init(isa_bus, pit);
1139
1140     for(i = 0; i < MAX_SERIAL_PORTS; i++) {
1141         if (serial_hds[i]) {
1142             serial_isa_init(isa_bus, i, serial_hds[i]);
1143         }
1144     }
1145
1146     for(i = 0; i < MAX_PARALLEL_PORTS; i++) {
1147         if (parallel_hds[i]) {
1148             parallel_init(isa_bus, i, parallel_hds[i]);
1149         }
1150     }
1151
1152     a20_line = qemu_allocate_irqs(handle_a20_line_change, first_cpu, 2);
1153     i8042 = isa_create_simple(isa_bus, "i8042");
1154     i8042_setup_a20_line(i8042, &a20_line[0]);
1155     if (!no_vmport) {
1156         vmport_init(isa_bus);
1157         vmmouse = isa_try_create(isa_bus, "vmmouse");
1158     } else {
1159         vmmouse = NULL;
1160     }
1161     if (vmmouse) {
1162         qdev_prop_set_ptr(&vmmouse->qdev, "ps2_mouse", i8042);
1163         qdev_init_nofail(&vmmouse->qdev);
1164     }
1165     port92 = isa_create_simple(isa_bus, "port92");
1166     port92_init(port92, &a20_line[1]);
1167
1168     cpu_exit_irq = qemu_allocate_irqs(cpu_request_exit, NULL, 1);
1169     DMA_init(0, cpu_exit_irq);
1170
1171     for(i = 0; i < MAX_FD; i++) {
1172         fd[i] = drive_get(IF_FLOPPY, 0, i);
1173     }
1174     *floppy = fdctrl_init_isa(isa_bus, fd);
1175 }
1176
1177 void pc_pci_device_init(PCIBus *pci_bus)
1178 {
1179     int max_bus;
1180     int bus;
1181
1182     max_bus = drive_get_max_bus(IF_SCSI);
1183     for (bus = 0; bus <= max_bus; bus++) {
1184         pci_create_simple(pci_bus, -1, "lsi53c895a");
1185     }
1186 }