tools/kvm/kvm.c

   1 #include "kvm/kvm.h"
   2
   3 #include "kvm/interrupt.h"
   4 #include "kvm/cpufeature.h"
   5 #include "kvm/e820.h"
   6 #include "kvm/util.h"
   7
   8 #include <linux/kvm.h>
   9
  10 #include <asm/bootparam.h>
  11
  12 #include <sys/ioctl.h>
  13 #include <inttypes.h>
  14 #include <sys/mman.h>
  15 #include <stdbool.h>
  16 #include <assert.h>
  17 #include <limits.h>
  18 #include <stdarg.h>
  19 #include <stdlib.h>
  20 #include <string.h>
  21 #include <unistd.h>
  22 #include <stdio.h>
  23 #include <fcntl.h>
  24 #include <sys/stat.h>
  25
  26 /*
  27  * Compatibility code. Remove this when we move to tools/kvm.
  28  */
  29 #ifndef KVM_EXIT_INTERNAL_ERROR
  30 # define KVM_EXIT_INTERNAL_ERROR                17
  31 #endif
  32
  33 #define DEFINE_KVM_EXIT_REASON(reason) [reason] = #reason
  34
  35 const char *kvm_exit_reasons[] = {
  36         DEFINE_KVM_EXIT_REASON(KVM_EXIT_UNKNOWN),
  37         DEFINE_KVM_EXIT_REASON(KVM_EXIT_EXCEPTION),
  38         DEFINE_KVM_EXIT_REASON(KVM_EXIT_IO),
  39         DEFINE_KVM_EXIT_REASON(KVM_EXIT_HYPERCALL),
  40         DEFINE_KVM_EXIT_REASON(KVM_EXIT_DEBUG),
  41         DEFINE_KVM_EXIT_REASON(KVM_EXIT_HLT),
  42         DEFINE_KVM_EXIT_REASON(KVM_EXIT_MMIO),
  43         DEFINE_KVM_EXIT_REASON(KVM_EXIT_IRQ_WINDOW_OPEN),
  44         DEFINE_KVM_EXIT_REASON(KVM_EXIT_SHUTDOWN),
  45         DEFINE_KVM_EXIT_REASON(KVM_EXIT_FAIL_ENTRY),
  46         DEFINE_KVM_EXIT_REASON(KVM_EXIT_INTR),
  47         DEFINE_KVM_EXIT_REASON(KVM_EXIT_SET_TPR),
  48         DEFINE_KVM_EXIT_REASON(KVM_EXIT_TPR_ACCESS),
  49         DEFINE_KVM_EXIT_REASON(KVM_EXIT_S390_SIEIC),
  50         DEFINE_KVM_EXIT_REASON(KVM_EXIT_S390_RESET),
  51         DEFINE_KVM_EXIT_REASON(KVM_EXIT_DCR),
  52         DEFINE_KVM_EXIT_REASON(KVM_EXIT_NMI),
  53         DEFINE_KVM_EXIT_REASON(KVM_EXIT_INTERNAL_ERROR),
  54 };
  55
  56 #define DEFINE_KVM_EXT(ext)             \
  57         .name = #ext,                   \
  58         .code = ext
  59
  60 struct {
  61         const char *name;
  62         int code;
  63 } kvm_req_ext[] = {
  64         { DEFINE_KVM_EXT(KVM_CAP_COALESCED_MMIO) },
  65         { DEFINE_KVM_EXT(KVM_CAP_SET_TSS_ADDR) },
  66         { DEFINE_KVM_EXT(KVM_CAP_PIT2) },
  67         { DEFINE_KVM_EXT(KVM_CAP_USER_MEMORY) },
  68         { DEFINE_KVM_EXT(KVM_CAP_IRQ_ROUTING) },
  69         { DEFINE_KVM_EXT(KVM_CAP_IRQCHIP) },
  70         { DEFINE_KVM_EXT(KVM_CAP_HLT) },
  71         { DEFINE_KVM_EXT(KVM_CAP_IRQ_INJECT_STATUS) },
  72         { DEFINE_KVM_EXT(KVM_CAP_EXT_CPUID) },
  73 };
  74
  75 static bool kvm__supports_extension(struct kvm *self, unsigned int extension)
  76 {
  77         int ret;
  78
  79         ret = ioctl(self->sys_fd, KVM_CHECK_EXTENSION, extension);
  80         if (ret < 0)
  81                 return false;
  82
  83         return ret;
  84 }
  85
  86 static int kvm__check_extensions(struct kvm *self)
  87 {
  88         unsigned int i;
  89
  90         for (i = 0; i < ARRAY_SIZE(kvm_req_ext); i++) {
  91                 if (!kvm__supports_extension(self, kvm_req_ext[i].code)) {
  92                         error("Unsuppored KVM extension detected: %s",
  93                                 kvm_req_ext[i].name);
  94                         return (int)-i;
  95                 }
  96         }
  97
  98         return 0;
  99 }
 100
 101 static struct kvm *kvm__new(void)
 102 {
 103         struct kvm *self = calloc(1, sizeof *self);
 104
 105         if (!self)
 106                 die("out of memory");
 107
 108         return self;
 109 }
 110
 111 void kvm__delete(struct kvm *self)
 112 {
 113         free(self->ram_start);
 114         free(self);
 115 }
 116
 117 static bool kvm__cpu_supports_vm(void)
 118 {
 119         struct cpuid_regs regs;
 120         uint32_t eax_base;
 121         int feature;
 122
 123         regs    = (struct cpuid_regs) {
 124                 .eax            = 0x00,
 125         };
 126         host_cpuid(&regs);
 127
 128         switch (regs.ebx) {
 129         case CPUID_VENDOR_INTEL_1:
 130                 eax_base        = 0x00;
 131                 feature         = KVM__X86_FEATURE_VMX;
 132                 break;
 133
 134         case CPUID_VENDOR_AMD_1:
 135                 eax_base        = 0x80000000;
 136                 feature         = KVM__X86_FEATURE_SVM;
 137                 break;
 138
 139         default:
 140                 return false;
 141         }
 142
 143         regs    = (struct cpuid_regs) {
 144                 .eax            = eax_base,
 145         };
 146         host_cpuid(&regs);
 147
 148         if (regs.eax < eax_base + 0x01)
 149                 return false;
 150
 151         regs    = (struct cpuid_regs) {
 152                 .eax            = eax_base + 0x01
 153         };
 154         host_cpuid(&regs);
 155
 156         return regs.ecx & (1 << feature);
 157 }
 158
 159 struct kvm *kvm__init(const char *kvm_dev, unsigned long ram_size)
 160 {
 161         struct kvm_userspace_memory_region mem;
 162         struct kvm_pit_config pit_config = { .flags = 0, };
 163         struct kvm *self;
 164         long page_size;
 165         int mmap_size;
 166         int ret;
 167
 168         if (!kvm__cpu_supports_vm())
 169                 die("Your CPU does not support hardware virtualization");
 170
 171         self = kvm__new();
 172
 173         self->sys_fd = open(kvm_dev, O_RDWR);
 174         if (self->sys_fd < 0) {
 175                 if (errno == ENOENT)
 176                         die("'%s' not found. Please make sure you have CONFIG_KVM enabled.", kvm_dev);
 177
 178                 die_perror("open");
 179         }
 180
 181         ret = ioctl(self->sys_fd, KVM_GET_API_VERSION, 0);
 182         if (ret != KVM_API_VERSION)
 183                 die_perror("KVM_API_VERSION ioctl");
 184
 185         self->vm_fd = ioctl(self->sys_fd, KVM_CREATE_VM, 0);
 186         if (self->vm_fd < 0)
 187                 die_perror("KVM_CREATE_VM ioctl");
 188
 189         if (kvm__check_extensions(self))
 190                 die("A required KVM extention is not supported by OS");
 191
 192         ret = ioctl(self->vm_fd, KVM_SET_TSS_ADDR, 0xfffbd000);
 193         if (ret < 0)
 194                 die_perror("KVM_SET_TSS_ADDR ioctl");
 195
 196         ret = ioctl(self->vm_fd, KVM_CREATE_PIT2, &pit_config);
 197         if (ret < 0)
 198                 die_perror("KVM_CREATE_PIT2 ioctl");
 199
 200         self->ram_size          = ram_size;
 201
 202         page_size       = sysconf(_SC_PAGESIZE);
 203         if (posix_memalign(&self->ram_start, page_size, self->ram_size) != 0)
 204                 die("out of memory");
 205
 206         mem = (struct kvm_userspace_memory_region) {
 207                 .slot                   = 0,
 208                 .guest_phys_addr        = 0x0UL,
 209                 .memory_size            = self->ram_size,
 210                 .userspace_addr         = (unsigned long) self->ram_start,
 211         };
 212
 213         ret = ioctl(self->vm_fd, KVM_SET_USER_MEMORY_REGION, &mem);
 214         if (ret < 0)
 215                 die_perror("KVM_SET_USER_MEMORY_REGION ioctl");
 216
 217         ret = ioctl(self->vm_fd, KVM_CREATE_IRQCHIP);
 218         if (ret < 0)
 219                 die_perror("KVM_CREATE_IRQCHIP ioctl");
 220
 221         self->vcpu_fd = ioctl(self->vm_fd, KVM_CREATE_VCPU, 0);
 222         if (self->vcpu_fd < 0)
 223                 die_perror("KVM_CREATE_VCPU ioctl");
 224
 225         mmap_size = ioctl(self->sys_fd, KVM_GET_VCPU_MMAP_SIZE, 0);
 226         if (mmap_size < 0)
 227                 die_perror("KVM_GET_VCPU_MMAP_SIZE ioctl");
 228
 229         self->kvm_run = mmap(NULL, mmap_size, PROT_READ|PROT_WRITE, MAP_SHARED, self->vcpu_fd, 0);
 230         if (self->kvm_run == MAP_FAILED)
 231                 die("unable to mmap vcpu fd");
 232
 233         return self;
 234 }
 235
 236 void kvm__enable_singlestep(struct kvm *self)
 237 {
 238         struct kvm_guest_debug debug = {
 239                 .control        = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP,
 240         };
 241
 242         if (ioctl(self->vcpu_fd, KVM_SET_GUEST_DEBUG, &debug) < 0)
 243                 warning("KVM_SET_GUEST_DEBUG failed");
 244 }
 245
 246 #define BOOT_LOADER_SELECTOR    0x1000
 247 #define BOOT_LOADER_IP          0x0000
 248 #define BOOT_LOADER_SP          0x8000
 249 #define BOOT_CMDLINE_OFFSET     0x20000
 250
 251 #define BOOT_PROTOCOL_REQUIRED  0x202
 252 #define LOAD_HIGH               0x01
 253
 254 static int load_flat_binary(struct kvm *self, int fd)
 255 {
 256         void *p;
 257         int nr;
 258
 259         if (lseek(fd, 0, SEEK_SET) < 0)
 260                 die_perror("lseek");
 261
 262         p = guest_real_to_host(self, BOOT_LOADER_SELECTOR, BOOT_LOADER_IP);
 263
 264         while ((nr = read(fd, p, 65536)) > 0)
 265                 p += nr;
 266
 267         self->boot_selector     = BOOT_LOADER_SELECTOR;
 268         self->boot_ip           = BOOT_LOADER_IP;
 269         self->boot_sp           = BOOT_LOADER_SP;
 270
 271         return true;
 272 }
 273
 274 /*
 275  * The protected mode kernel part of a modern bzImage is loaded at 1 MB by
 276  * default.
 277  */
 278 #define BZ_KERNEL_START                 0x100000UL
 279 #define INITRD_START                    0x1000000UL
 280 #define BZ_DEFAULT_SETUP_SECTS          4
 281 static const char *BZIMAGE_MAGIC        = "HdrS";
 282
 283 static bool load_bzimage(struct kvm *self, int fd_kernel,
 284                         int fd_initrd, const char *kernel_cmdline)
 285 {
 286         struct boot_params *kern_boot;
 287         unsigned long setup_sects;
 288         struct boot_params boot;
 289         size_t cmdline_size;
 290         ssize_t setup_size;
 291         void *p;
 292         int nr;
 293
 294         /*
 295          * See Documentation/x86/boot.txt for details no bzImage on-disk and
 296          * memory layout.
 297          */
 298
 299         if (lseek(fd_kernel, 0, SEEK_SET) < 0)
 300                 die_perror("lseek");
 301
 302         if (read(fd_kernel, &boot, sizeof(boot)) != sizeof(boot))
 303                 return false;
 304
 305         if (memcmp(&boot.hdr.header, BZIMAGE_MAGIC, strlen(BZIMAGE_MAGIC)))
 306                 return false;
 307
 308         if (boot.hdr.version < BOOT_PROTOCOL_REQUIRED) {
 309                 die("Too old kernel");
 310         }
 311
 312         if (lseek(fd_kernel, 0, SEEK_SET) < 0)
 313                 die_perror("lseek");
 314
 315         if (!boot.hdr.setup_sects)
 316                 boot.hdr.setup_sects = BZ_DEFAULT_SETUP_SECTS;
 317         setup_sects = boot.hdr.setup_sects + 1;
 318
 319         setup_size = setup_sects << 9;
 320         p = guest_real_to_host(self, BOOT_LOADER_SELECTOR, BOOT_LOADER_IP);
 321
 322         /* copy setup.bin to mem*/
 323         if (read(fd_kernel, p, setup_size) != setup_size)
 324                 die_perror("read");
 325
 326         /* copy vmlinux.bin to BZ_KERNEL_START*/
 327         p = guest_flat_to_host(self, BZ_KERNEL_START);
 328
 329         while ((nr = read(fd_kernel, p, 65536)) > 0)
 330                 p += nr;
 331
 332         p = guest_flat_to_host(self, BOOT_CMDLINE_OFFSET);
 333         if (kernel_cmdline) {
 334                 cmdline_size = strlen(kernel_cmdline) + 1;
 335                 if (cmdline_size > boot.hdr.cmdline_size)
 336                         cmdline_size = boot.hdr.cmdline_size;
 337
 338                 memset(p, 0, boot.hdr.cmdline_size);
 339                 memcpy(p, kernel_cmdline, cmdline_size - 1);
 340         }
 341
 342         kern_boot       = guest_real_to_host(self, BOOT_LOADER_SELECTOR, 0x00);
 343
 344         kern_boot->hdr.cmd_line_ptr     = BOOT_CMDLINE_OFFSET;
 345         kern_boot->hdr.type_of_loader   = 0xff;
 346         kern_boot->hdr.heap_end_ptr     = 0xfe00;
 347         kern_boot->hdr.loadflags        |= CAN_USE_HEAP;
 348
 349         /*
 350          * Read initrd image into guest memory
 351          */
 352         if (fd_initrd >= 0) {
 353                 struct stat initrd_stat;
 354                 unsigned long addr;
 355
 356                 if (fstat(fd_initrd, &initrd_stat))
 357                         die_perror("fstat");
 358
 359                 addr = boot.hdr.initrd_addr_max & ~0xfffff;
 360                 for (;;) {
 361                         if (addr < BZ_KERNEL_START)
 362                                 die("Not enough memory for initrd");
 363                         else if (addr < (self->ram_size - initrd_stat.st_size))
 364                                 break;
 365                         addr -= 0x100000;
 366                 }
 367
 368                 p = guest_flat_to_host(self, addr);
 369                 nr = read(fd_initrd, p, initrd_stat.st_size);
 370                 if (nr != initrd_stat.st_size)
 371                         die("Failed to read initrd");
 372
 373                 kern_boot->hdr.ramdisk_image    = addr;
 374                 kern_boot->hdr.ramdisk_size     = initrd_stat.st_size;
 375         }
 376
 377         self->boot_selector     = BOOT_LOADER_SELECTOR;
 378         /*
 379          * The real-mode setup code starts at offset 0x200 of a bzImage. See
 380          * Documentation/x86/boot.txt for details.
 381          */
 382         self->boot_ip           = BOOT_LOADER_IP + 0x200;
 383         self->boot_sp           = BOOT_LOADER_SP;
 384
 385         /*
 386          * Drum roll, BIOS is coming to live, oh dear...
 387          */
 388         setup_bios(self);
 389
 390         return true;
 391 }
 392
 393 bool kvm__load_kernel(struct kvm *kvm, const char *kernel_filename,
 394                 const char *initrd_filename, const char *kernel_cmdline)
 395 {
 396         bool ret;
 397         int fd_kernel = -1, fd_initrd = -1;
 398
 399         fd_kernel = open(kernel_filename, O_RDONLY);
 400         if (fd_kernel < 0)
 401                 die("Unable to open kernel %s", kernel_filename);
 402
 403         if (initrd_filename) {
 404                 fd_initrd = open(initrd_filename, O_RDONLY);
 405                 if (fd_initrd < 0)
 406                         die("Unable to open initrd %s", initrd_filename);
 407         }
 408
 409         ret = load_bzimage(kvm, fd_kernel, fd_initrd, kernel_cmdline);
 410
 411         if (initrd_filename)
 412                 close(fd_initrd);
 413
 414         if (ret)
 415                 goto found_kernel;
 416
 417         warning("%s is not a bzImage. Trying to load it as a flat binary...", kernel_filename);
 418
 419         ret = load_flat_binary(kvm, fd_kernel);
 420         if (ret)
 421                 goto found_kernel;
 422
 423         die("%s is not a valid bzImage or flat binary", kernel_filename);
 424
 425 found_kernel:
 426         return ret;
 427 }
 428
 429 static inline uint64_t ip_flat_to_real(struct kvm *self, uint64_t ip)
 430 {
 431         uint64_t cs = self->sregs.cs.selector;
 432
 433         return ip - (cs << 4);
 434 }
 435
 436 static inline bool is_in_protected_mode(struct kvm *self)
 437 {
 438         return self->sregs.cr0 & 0x01;
 439 }
 440
 441 static inline uint64_t ip_to_flat(struct kvm *self, uint64_t ip)
 442 {
 443         uint64_t cs;
 444
 445         /*
 446          * NOTE! We should take code segment base address into account here.
 447          * Luckily it's usually zero because Linux uses flat memory model.
 448          */
 449         if (is_in_protected_mode(self))
 450                 return ip;
 451
 452         cs = self->sregs.cs.selector;
 453
 454         return ip + (cs << 4);
 455 }
 456
 457 static inline uint32_t selector_to_base(uint16_t selector)
 458 {
 459         /*
 460          * KVM on Intel requires 'base' to be 'selector * 16' in real mode.
 461          */
 462         return (uint32_t)selector * 16;
 463 }
 464
 465 static struct kvm_msrs *kvm_msrs__new(size_t nmsrs)
 466 {
 467         struct kvm_msrs *self = calloc(1, sizeof(*self) + (sizeof(struct kvm_msr_entry) * nmsrs));
 468
 469         if (!self)
 470                 die("out of memory");
 471
 472         return self;
 473 }
 474
 475 #define MSR_IA32_TIME_STAMP_COUNTER     0x10
 476
 477 #define MSR_IA32_SYSENTER_CS            0x174
 478 #define MSR_IA32_SYSENTER_ESP           0x175
 479 #define MSR_IA32_SYSENTER_EIP           0x176
 480
 481 #define MSR_IA32_STAR                   0xc0000081
 482 #define MSR_IA32_LSTAR                  0xc0000082
 483 #define MSR_IA32_CSTAR                  0xc0000083
 484 #define MSR_IA32_FMASK                  0xc0000084
 485 #define MSR_IA32_KERNEL_GS_BASE         0xc0000102
 486
 487 #define KVM_MSR_ENTRY(_index, _data)    \
 488         (struct kvm_msr_entry) { .index = _index, .data = _data }
 489
 490 static void kvm__setup_msrs(struct kvm *self)
 491 {
 492         unsigned long ndx = 0;
 493
 494         self->msrs = kvm_msrs__new(100);
 495
 496         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_CS,        0x0);
 497         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_ESP,       0x0);
 498         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_EIP,       0x0);
 499 #ifdef CONFIG_X86_64
 500         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_STAR,               0x0);
 501         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_CSTAR,              0x0);
 502         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_KERNEL_GS_BASE,     0x0);
 503         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_FMASK,              0x0);
 504         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_LSTAR,              0x0);
 505 #endif
 506         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_TIME_STAMP_COUNTER, 0x0);
 507
 508         self->msrs->nmsrs       = ndx;
 509
 510         if (ioctl(self->vcpu_fd, KVM_SET_MSRS, self->msrs) < 0)
 511                 die_perror("KVM_SET_MSRS failed");
 512 }
 513
 514 static void kvm__setup_fpu(struct kvm *self)
 515 {
 516         self->fpu = (struct kvm_fpu) {
 517                 .fcw            = 0x37f,
 518                 .mxcsr          = 0x1f80,
 519         };
 520
 521         if (ioctl(self->vcpu_fd, KVM_SET_FPU, &self->fpu) < 0)
 522                 die_perror("KVM_SET_FPU failed");
 523 }
 524
 525 static void kvm__setup_regs(struct kvm *self)
 526 {
 527         self->regs = (struct kvm_regs) {
 528                 /* We start the guest in 16-bit real mode  */
 529                 .rflags         = 0x0000000000000002ULL,
 530
 531                 .rip            = self->boot_ip,
 532                 .rsp            = self->boot_sp,
 533                 .rbp            = self->boot_sp,
 534         };
 535
 536         if (self->regs.rip > USHRT_MAX)
 537                 die("ip 0x%" PRIx64 " is too high for real mode", (uint64_t) self->regs.rip);
 538
 539         if (ioctl(self->vcpu_fd, KVM_SET_REGS, &self->regs) < 0)
 540                 die_perror("KVM_SET_REGS failed");
 541 }
 542
 543 static void kvm__setup_sregs(struct kvm *self)
 544 {
 545
 546         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &self->sregs) < 0)
 547                 die_perror("KVM_GET_SREGS failed");
 548
 549         self->sregs.cs.selector = self->boot_selector;
 550         self->sregs.cs.base     = selector_to_base(self->boot_selector);
 551         self->sregs.ss.selector = self->boot_selector;
 552         self->sregs.ss.base     = selector_to_base(self->boot_selector);
 553         self->sregs.ds.selector = self->boot_selector;
 554         self->sregs.ds.base     = selector_to_base(self->boot_selector);
 555         self->sregs.es.selector = self->boot_selector;
 556         self->sregs.es.base     = selector_to_base(self->boot_selector);
 557         self->sregs.fs.selector = self->boot_selector;
 558         self->sregs.fs.base     = selector_to_base(self->boot_selector);
 559         self->sregs.gs.selector = self->boot_selector;
 560         self->sregs.gs.base     = selector_to_base(self->boot_selector);
 561
 562         if (ioctl(self->vcpu_fd, KVM_SET_SREGS, &self->sregs) < 0)
 563                 die_perror("KVM_SET_SREGS failed");
 564 }
 565
 566 void kvm__reset_vcpu(struct kvm *self)
 567 {
 568         kvm__setup_sregs(self);
 569
 570         kvm__setup_regs(self);
 571
 572         kvm__setup_fpu(self);
 573
 574         kvm__setup_msrs(self);
 575 }
 576
 577 void kvm__setup_mem(struct kvm *self)
 578 {
 579         struct e820_entry *mem_map;
 580         unsigned char *size;
 581
 582         size            = guest_flat_to_host(self, E820_MAP_SIZE);
 583         mem_map         = guest_flat_to_host(self, E820_MAP_START);
 584
 585         *size           = 4;
 586
 587         mem_map[0]      = (struct e820_entry) {
 588                 .addr           = REAL_MODE_IVT_BEGIN,
 589                 .size           = EBDA_START - REAL_MODE_IVT_BEGIN,
 590                 .type           = E820_MEM_USABLE,
 591         };
 592         mem_map[1]      = (struct e820_entry) {
 593                 .addr           = EBDA_START,
 594                 .size           = VGA_RAM_BEGIN - EBDA_START,
 595                 .type           = E820_MEM_RESERVED,
 596         };
 597         mem_map[2]      = (struct e820_entry) {
 598                 .addr           = MB_BIOS_BEGIN,
 599                 .size           = MB_BIOS_END - MB_BIOS_BEGIN,
 600                 .type           = E820_MEM_RESERVED,
 601         };
 602         mem_map[3]      = (struct e820_entry) {
 603                 .addr           = BZ_KERNEL_START,
 604                 .size           = self->ram_size - BZ_KERNEL_START,
 605                 .type           = E820_MEM_USABLE,
 606         };
 607 }
 608
 609 void kvm__run(struct kvm *self)
 610 {
 611         if (ioctl(self->vcpu_fd, KVM_RUN, 0) < 0)
 612                 die_perror("KVM_RUN failed");
 613 }
 614
 615 void kvm__irq_line(struct kvm *self, int irq, int level)
 616 {
 617         struct kvm_irq_level irq_level;
 618
 619         irq_level       = (struct kvm_irq_level) {
 620                 {
 621                         .irq            = irq,
 622                 },
 623                 .level          = level,
 624         };
 625
 626         if (ioctl(self->vm_fd, KVM_IRQ_LINE, &irq_level) < 0)
 627                 die_perror("KVM_IRQ_LINE failed");
 628 }
 629
 630 static void print_dtable(const char *name, struct kvm_dtable *dtable)
 631 {
 632         printf(" %s                 %016" PRIx64 "  %08" PRIx16 "\n",
 633                 name, (uint64_t) dtable->base, (uint16_t) dtable->limit);
 634 }
 635
 636 static void print_segment(const char *name, struct kvm_segment *seg)
 637 {
 638         printf(" %s       %04" PRIx16 "      %016" PRIx64 "  %08" PRIx32 "  %02" PRIx8 "    %x %x   %x  %x %x %x %x\n",
 639                 name, (uint16_t) seg->selector, (uint64_t) seg->base, (uint32_t) seg->limit,
 640                 (uint8_t) seg->type, seg->present, seg->dpl, seg->db, seg->s, seg->l, seg->g, seg->avl);
 641 }
 642
 643 void kvm__show_registers(struct kvm *self)
 644 {
 645         unsigned long cr0, cr2, cr3;
 646         unsigned long cr4, cr8;
 647         unsigned long rax, rbx, rcx;
 648         unsigned long rdx, rsi, rdi;
 649         unsigned long rbp,  r8,  r9;
 650         unsigned long r10, r11, r12;
 651         unsigned long r13, r14, r15;
 652         unsigned long rip, rsp;
 653         struct kvm_sregs sregs;
 654         unsigned long rflags;
 655         struct kvm_regs regs;
 656         int i;
 657
 658         if (ioctl(self->vcpu_fd, KVM_GET_REGS, &regs) < 0)
 659                 die("KVM_GET_REGS failed");
 660
 661         rflags = regs.rflags;
 662
 663         rip = regs.rip; rsp = regs.rsp;
 664         rax = regs.rax; rbx = regs.rbx; rcx = regs.rcx;
 665         rdx = regs.rdx; rsi = regs.rsi; rdi = regs.rdi;
 666         rbp = regs.rbp; r8  = regs.r8;  r9  = regs.r9;
 667         r10 = regs.r10; r11 = regs.r11; r12 = regs.r12;
 668         r13 = regs.r13; r14 = regs.r14; r15 = regs.r15;
 669
 670         printf("Registers:\n");
 671         printf(" rip: %016lx   rsp: %016lx flags: %016lx\n", rip, rsp, rflags);
 672         printf(" rax: %016lx   rbx: %016lx   rcx: %016lx\n", rax, rbx, rcx);
 673         printf(" rdx: %016lx   rsi: %016lx   rdi: %016lx\n", rdx, rsi, rdi);
 674         printf(" rbp: %016lx   r8:  %016lx   r9:  %016lx\n", rbp, r8,  r9);
 675         printf(" r10: %016lx   r11: %016lx   r12: %016lx\n", r10, r11, r12);
 676         printf(" r13: %016lx   r14: %016lx   r15: %016lx\n", r13, r14, r15);
 677
 678         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &sregs) < 0)
 679                 die("KVM_GET_REGS failed");
 680
 681         cr0 = sregs.cr0; cr2 = sregs.cr2; cr3 = sregs.cr3;
 682         cr4 = sregs.cr4; cr8 = sregs.cr8;
 683
 684         printf(" cr0: %016lx   cr2: %016lx   cr3: %016lx\n", cr0, cr2, cr3);
 685         printf(" cr4: %016lx   cr8: %016lx\n", cr4, cr8);
 686         printf("Segment registers:\n");
 687         printf(" register  selector  base              limit     type  p dpl db s l g avl\n");
 688         print_segment("cs ", &sregs.cs);
 689         print_segment("ss ", &sregs.ss);
 690         print_segment("ds ", &sregs.ds);
 691         print_segment("es ", &sregs.es);
 692         print_segment("fs ", &sregs.fs);
 693         print_segment("gs ", &sregs.gs);
 694         print_segment("tr ", &sregs.tr);
 695         print_segment("ldt", &sregs.ldt);
 696         print_dtable("gdt", &sregs.gdt);
 697         print_dtable("idt", &sregs.idt);
 698         printf(" [ efer: %016" PRIx64 "  apic base: %016" PRIx64 "  nmi: %s ]\n",
 699                 (uint64_t) sregs.efer, (uint64_t) sregs.apic_base,
 700                 (self->nmi_disabled ? "disabled" : "enabled"));
 701         printf("Interrupt bitmap:\n");
 702         printf(" ");
 703         for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++)
 704                 printf("%016" PRIx64 " ", (uint64_t) sregs.interrupt_bitmap[i]);
 705         printf("\n");
 706 }
 707
 708 void kvm__show_code(struct kvm *self)
 709 {
 710         unsigned int code_bytes = 64;
 711         unsigned int code_prologue = code_bytes * 43 / 64;
 712         unsigned int code_len = code_bytes;
 713         unsigned char c;
 714         unsigned int i;
 715         uint8_t *ip;
 716
 717         if (ioctl(self->vcpu_fd, KVM_GET_REGS, &self->regs) < 0)
 718                 die("KVM_GET_REGS failed");
 719
 720         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &self->sregs) < 0)
 721                 die("KVM_GET_SREGS failed");
 722
 723         ip = guest_flat_to_host(self, ip_to_flat(self, self->regs.rip) - code_prologue);
 724
 725         printf("Code: ");
 726
 727         for (i = 0; i < code_len; i++, ip++) {
 728                 if (!host_ptr_in_ram(self, ip))
 729                         break;
 730
 731                 c = *ip;
 732
 733                 if (ip == guest_flat_to_host(self, ip_to_flat(self, self->regs.rip)))
 734                         printf("<%02x> ", c);
 735                 else
 736                         printf("%02x ", c);
 737         }
 738
 739         printf("\n");
 740
 741         printf("Stack:\n");
 742         kvm__dump_mem(self, self->regs.rsp, 32);
 743 }
 744
 745 void kvm__show_page_tables(struct kvm *self)
 746 {
 747         uint64_t *pte1;
 748         uint64_t *pte2;
 749         uint64_t *pte3;
 750         uint64_t *pte4;
 751
 752         if (!is_in_protected_mode(self))
 753                 return;
 754
 755         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &self->sregs) < 0)
 756                 die("KVM_GET_SREGS failed");
 757
 758         pte4    = guest_flat_to_host(self, self->sregs.cr3);
 759         if (!host_ptr_in_ram(self, pte4))
 760                 return;
 761
 762         pte3    = guest_flat_to_host(self, (*pte4 & ~0xfff));
 763         if (!host_ptr_in_ram(self, pte3))
 764                 return;
 765
 766         pte2    = guest_flat_to_host(self, (*pte3 & ~0xfff));
 767         if (!host_ptr_in_ram(self, pte2))
 768                 return;
 769
 770         pte1    = guest_flat_to_host(self, (*pte2 & ~0xfff));
 771         if (!host_ptr_in_ram(self, pte1))
 772                 return;
 773
 774         printf("Page Tables:\n");
 775         if (*pte2 & (1 << 7))
 776                 printf(" pte4: %016" PRIx64 "   pte3: %016" PRIx64
 777                         "   pte2: %016" PRIx64 "\n",
 778                         *pte4, *pte3, *pte2);
 779         else
 780                 printf(" pte4: %016" PRIx64 "   pte3: %016" PRIx64 "   pte2: %016"
 781                         PRIx64 "   pte1: %016" PRIx64 "\n",
 782                         *pte4, *pte3, *pte2, *pte1);
 783 }
 784
 785 void kvm__dump_mem(struct kvm *self, unsigned long addr, unsigned long size)
 786 {
 787         unsigned char *p;
 788         unsigned long n;
 789
 790         size &= ~7; /* mod 8 */
 791         if (!size)
 792                 return;
 793
 794         p = guest_flat_to_host(self, addr);
 795
 796         for (n = 0; n < size; n+=8) {
 797                 if (!host_ptr_in_ram(self, p + n))
 798                         break;
 799
 800                 printf("  0x%08lx: %02x %02x %02x %02x  %02x %02x %02x %02x\n",
 801                         addr + n, p[n + 0], p[n + 1], p[n + 2], p[n + 3],
 802                                   p[n + 4], p[n + 5], p[n + 6], p[n + 7]);
 803         }
 804 }