tools/kvm/kvm.c

   1 #include "kvm/kvm.h"
   2
   3 #include "kvm/cpufeature.h"
   4 #include "kvm/interrupt.h"
   5 #include "kvm/boot-protocol.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 <sys/stat.h>
  16 #include <stdbool.h>
  17 #include <assert.h>
  18 #include <limits.h>
  19 #include <signal.h>
  20 #include <stdarg.h>
  21 #include <stdlib.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <stdio.h>
  25 #include <fcntl.h>
  26 #include <time.h>
  27
  28 #define DEFINE_KVM_EXIT_REASON(reason) [reason] = #reason
  29
  30 const char *kvm_exit_reasons[] = {
  31         DEFINE_KVM_EXIT_REASON(KVM_EXIT_UNKNOWN),
  32         DEFINE_KVM_EXIT_REASON(KVM_EXIT_EXCEPTION),
  33         DEFINE_KVM_EXIT_REASON(KVM_EXIT_IO),
  34         DEFINE_KVM_EXIT_REASON(KVM_EXIT_HYPERCALL),
  35         DEFINE_KVM_EXIT_REASON(KVM_EXIT_DEBUG),
  36         DEFINE_KVM_EXIT_REASON(KVM_EXIT_HLT),
  37         DEFINE_KVM_EXIT_REASON(KVM_EXIT_MMIO),
  38         DEFINE_KVM_EXIT_REASON(KVM_EXIT_IRQ_WINDOW_OPEN),
  39         DEFINE_KVM_EXIT_REASON(KVM_EXIT_SHUTDOWN),
  40         DEFINE_KVM_EXIT_REASON(KVM_EXIT_FAIL_ENTRY),
  41         DEFINE_KVM_EXIT_REASON(KVM_EXIT_INTR),
  42         DEFINE_KVM_EXIT_REASON(KVM_EXIT_SET_TPR),
  43         DEFINE_KVM_EXIT_REASON(KVM_EXIT_TPR_ACCESS),
  44         DEFINE_KVM_EXIT_REASON(KVM_EXIT_S390_SIEIC),
  45         DEFINE_KVM_EXIT_REASON(KVM_EXIT_S390_RESET),
  46         DEFINE_KVM_EXIT_REASON(KVM_EXIT_DCR),
  47         DEFINE_KVM_EXIT_REASON(KVM_EXIT_NMI),
  48         DEFINE_KVM_EXIT_REASON(KVM_EXIT_INTERNAL_ERROR),
  49 };
  50
  51 #define DEFINE_KVM_EXT(ext)             \
  52         .name = #ext,                   \
  53         .code = ext
  54
  55 struct {
  56         const char *name;
  57         int code;
  58 } kvm_req_ext[] = {
  59         { DEFINE_KVM_EXT(KVM_CAP_COALESCED_MMIO) },
  60         { DEFINE_KVM_EXT(KVM_CAP_SET_TSS_ADDR) },
  61         { DEFINE_KVM_EXT(KVM_CAP_PIT2) },
  62         { DEFINE_KVM_EXT(KVM_CAP_USER_MEMORY) },
  63         { DEFINE_KVM_EXT(KVM_CAP_IRQ_ROUTING) },
  64         { DEFINE_KVM_EXT(KVM_CAP_IRQCHIP) },
  65         { DEFINE_KVM_EXT(KVM_CAP_HLT) },
  66         { DEFINE_KVM_EXT(KVM_CAP_IRQ_INJECT_STATUS) },
  67         { DEFINE_KVM_EXT(KVM_CAP_EXT_CPUID) },
  68 };
  69
  70 static bool kvm__supports_extension(struct kvm *self, unsigned int extension)
  71 {
  72         int ret;
  73
  74         ret = ioctl(self->sys_fd, KVM_CHECK_EXTENSION, extension);
  75         if (ret < 0)
  76                 return false;
  77
  78         return ret;
  79 }
  80
  81 static int kvm__check_extensions(struct kvm *self)
  82 {
  83         unsigned int i;
  84
  85         for (i = 0; i < ARRAY_SIZE(kvm_req_ext); i++) {
  86                 if (!kvm__supports_extension(self, kvm_req_ext[i].code)) {
  87                         error("Unsuppored KVM extension detected: %s",
  88                                 kvm_req_ext[i].name);
  89                         return (int)-i;
  90                 }
  91         }
  92
  93         return 0;
  94 }
  95
  96 static struct kvm *kvm__new(void)
  97 {
  98         struct kvm *self = calloc(1, sizeof *self);
  99
 100         if (!self)
 101                 die("out of memory");
 102
 103         return self;
 104 }
 105
 106 void kvm__delete(struct kvm *self)
 107 {
 108         kvm__stop_timer(self);
 109
 110         if (self->msrs)
 111                 free(self->msrs);
 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 your kernel has CONFIG_KVM enabled and that the KVM modules are loaded.", 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  0x206
 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 static const char *BZIMAGE_MAGIC        = "HdrS";
 275
 276 static bool load_bzimage(struct kvm *self, int fd_kernel,
 277                         int fd_initrd, const char *kernel_cmdline)
 278 {
 279         struct boot_params *kern_boot;
 280         unsigned long setup_sects;
 281         struct boot_params boot;
 282         size_t cmdline_size;
 283         ssize_t setup_size;
 284         void *p;
 285         int nr;
 286
 287         /*
 288          * See Documentation/x86/boot.txt for details no bzImage on-disk and
 289          * memory layout.
 290          */
 291
 292         if (lseek(fd_kernel, 0, SEEK_SET) < 0)
 293                 die_perror("lseek");
 294
 295         if (read(fd_kernel, &boot, sizeof(boot)) != sizeof(boot))
 296                 return false;
 297
 298         if (memcmp(&boot.hdr.header, BZIMAGE_MAGIC, strlen(BZIMAGE_MAGIC)))
 299                 return false;
 300
 301         if (boot.hdr.version < BOOT_PROTOCOL_REQUIRED) {
 302                 die("Too old kernel");
 303         }
 304
 305         if (lseek(fd_kernel, 0, SEEK_SET) < 0)
 306                 die_perror("lseek");
 307
 308         if (!boot.hdr.setup_sects)
 309                 boot.hdr.setup_sects = BZ_DEFAULT_SETUP_SECTS;
 310         setup_sects = boot.hdr.setup_sects + 1;
 311
 312         setup_size = setup_sects << 9;
 313         p = guest_real_to_host(self, BOOT_LOADER_SELECTOR, BOOT_LOADER_IP);
 314
 315         /* copy setup.bin to mem*/
 316         if (read(fd_kernel, p, setup_size) != setup_size)
 317                 die_perror("read");
 318
 319         /* copy vmlinux.bin to BZ_KERNEL_START*/
 320         p = guest_flat_to_host(self, BZ_KERNEL_START);
 321
 322         while ((nr = read(fd_kernel, p, 65536)) > 0)
 323                 p += nr;
 324
 325         p = guest_flat_to_host(self, BOOT_CMDLINE_OFFSET);
 326         if (kernel_cmdline) {
 327                 cmdline_size = strlen(kernel_cmdline) + 1;
 328                 if (cmdline_size > boot.hdr.cmdline_size)
 329                         cmdline_size = boot.hdr.cmdline_size;
 330
 331                 memset(p, 0, boot.hdr.cmdline_size);
 332                 memcpy(p, kernel_cmdline, cmdline_size - 1);
 333         }
 334
 335         kern_boot       = guest_real_to_host(self, BOOT_LOADER_SELECTOR, 0x00);
 336
 337         kern_boot->hdr.cmd_line_ptr     = BOOT_CMDLINE_OFFSET;
 338         kern_boot->hdr.type_of_loader   = 0xff;
 339         kern_boot->hdr.heap_end_ptr     = 0xfe00;
 340         kern_boot->hdr.loadflags        |= CAN_USE_HEAP;
 341
 342         /*
 343          * Read initrd image into guest memory
 344          */
 345         if (fd_initrd >= 0) {
 346                 struct stat initrd_stat;
 347                 unsigned long addr;
 348
 349                 if (fstat(fd_initrd, &initrd_stat))
 350                         die_perror("fstat");
 351
 352                 addr = boot.hdr.initrd_addr_max & ~0xfffff;
 353                 for (;;) {
 354                         if (addr < BZ_KERNEL_START)
 355                                 die("Not enough memory for initrd");
 356                         else if (addr < (self->ram_size - initrd_stat.st_size))
 357                                 break;
 358                         addr -= 0x100000;
 359                 }
 360
 361                 p = guest_flat_to_host(self, addr);
 362                 nr = read(fd_initrd, p, initrd_stat.st_size);
 363                 if (nr != initrd_stat.st_size)
 364                         die("Failed to read initrd");
 365
 366                 kern_boot->hdr.ramdisk_image    = addr;
 367                 kern_boot->hdr.ramdisk_size     = initrd_stat.st_size;
 368         }
 369
 370         self->boot_selector     = BOOT_LOADER_SELECTOR;
 371         /*
 372          * The real-mode setup code starts at offset 0x200 of a bzImage. See
 373          * Documentation/x86/boot.txt for details.
 374          */
 375         self->boot_ip           = BOOT_LOADER_IP + 0x200;
 376         self->boot_sp           = BOOT_LOADER_SP;
 377
 378         return true;
 379 }
 380
 381 bool kvm__load_kernel(struct kvm *kvm, const char *kernel_filename,
 382                 const char *initrd_filename, const char *kernel_cmdline)
 383 {
 384         bool ret;
 385         int fd_kernel = -1, fd_initrd = -1;
 386
 387         fd_kernel = open(kernel_filename, O_RDONLY);
 388         if (fd_kernel < 0)
 389                 die("Unable to open kernel %s", kernel_filename);
 390
 391         if (initrd_filename) {
 392                 fd_initrd = open(initrd_filename, O_RDONLY);
 393                 if (fd_initrd < 0)
 394                         die("Unable to open initrd %s", initrd_filename);
 395         }
 396
 397         ret = load_bzimage(kvm, fd_kernel, fd_initrd, kernel_cmdline);
 398
 399         if (initrd_filename)
 400                 close(fd_initrd);
 401
 402         if (ret)
 403                 goto found_kernel;
 404
 405         warning("%s is not a bzImage. Trying to load it as a flat binary...", kernel_filename);
 406
 407         ret = load_flat_binary(kvm, fd_kernel);
 408         if (ret)
 409                 goto found_kernel;
 410
 411         die("%s is not a valid bzImage or flat binary", kernel_filename);
 412
 413 found_kernel:
 414         return ret;
 415 }
 416
 417 static inline uint64_t ip_flat_to_real(struct kvm *self, uint64_t ip)
 418 {
 419         uint64_t cs = self->sregs.cs.selector;
 420
 421         return ip - (cs << 4);
 422 }
 423
 424 static inline bool is_in_protected_mode(struct kvm *self)
 425 {
 426         return self->sregs.cr0 & 0x01;
 427 }
 428
 429 static inline uint64_t ip_to_flat(struct kvm *self, uint64_t ip)
 430 {
 431         uint64_t cs;
 432
 433         /*
 434          * NOTE! We should take code segment base address into account here.
 435          * Luckily it's usually zero because Linux uses flat memory model.
 436          */
 437         if (is_in_protected_mode(self))
 438                 return ip;
 439
 440         cs = self->sregs.cs.selector;
 441
 442         return ip + (cs << 4);
 443 }
 444
 445 static inline uint32_t selector_to_base(uint16_t selector)
 446 {
 447         /*
 448          * KVM on Intel requires 'base' to be 'selector * 16' in real mode.
 449          */
 450         return (uint32_t)selector * 16;
 451 }
 452
 453 static struct kvm_msrs *kvm_msrs__new(size_t nmsrs)
 454 {
 455         struct kvm_msrs *self = calloc(1, sizeof(*self) + (sizeof(struct kvm_msr_entry) * nmsrs));
 456
 457         if (!self)
 458                 die("out of memory");
 459
 460         return self;
 461 }
 462
 463 #define MSR_IA32_TIME_STAMP_COUNTER     0x10
 464
 465 #define MSR_IA32_SYSENTER_CS            0x174
 466 #define MSR_IA32_SYSENTER_ESP           0x175
 467 #define MSR_IA32_SYSENTER_EIP           0x176
 468
 469 #define MSR_IA32_STAR                   0xc0000081
 470 #define MSR_IA32_LSTAR                  0xc0000082
 471 #define MSR_IA32_CSTAR                  0xc0000083
 472 #define MSR_IA32_FMASK                  0xc0000084
 473 #define MSR_IA32_KERNEL_GS_BASE         0xc0000102
 474
 475 #define KVM_MSR_ENTRY(_index, _data)    \
 476         (struct kvm_msr_entry) { .index = _index, .data = _data }
 477
 478 static void kvm__setup_msrs(struct kvm *self)
 479 {
 480         unsigned long ndx = 0;
 481
 482         self->msrs = kvm_msrs__new(100);
 483
 484         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_CS,        0x0);
 485         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_ESP,       0x0);
 486         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_EIP,       0x0);
 487 #ifdef CONFIG_X86_64
 488         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_STAR,               0x0);
 489         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_CSTAR,              0x0);
 490         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_KERNEL_GS_BASE,     0x0);
 491         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_FMASK,              0x0);
 492         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_LSTAR,              0x0);
 493 #endif
 494         self->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_TIME_STAMP_COUNTER, 0x0);
 495
 496         self->msrs->nmsrs       = ndx;
 497
 498         if (ioctl(self->vcpu_fd, KVM_SET_MSRS, self->msrs) < 0)
 499                 die_perror("KVM_SET_MSRS failed");
 500 }
 501
 502 static void kvm__setup_fpu(struct kvm *self)
 503 {
 504         self->fpu = (struct kvm_fpu) {
 505                 .fcw            = 0x37f,
 506                 .mxcsr          = 0x1f80,
 507         };
 508
 509         if (ioctl(self->vcpu_fd, KVM_SET_FPU, &self->fpu) < 0)
 510                 die_perror("KVM_SET_FPU failed");
 511 }
 512
 513 static void kvm__setup_regs(struct kvm *self)
 514 {
 515         self->regs = (struct kvm_regs) {
 516                 /* We start the guest in 16-bit real mode  */
 517                 .rflags         = 0x0000000000000002ULL,
 518
 519                 .rip            = self->boot_ip,
 520                 .rsp            = self->boot_sp,
 521                 .rbp            = self->boot_sp,
 522         };
 523
 524         if (self->regs.rip > USHRT_MAX)
 525                 die("ip 0x%" PRIx64 " is too high for real mode", (uint64_t) self->regs.rip);
 526
 527         if (ioctl(self->vcpu_fd, KVM_SET_REGS, &self->regs) < 0)
 528                 die_perror("KVM_SET_REGS failed");
 529 }
 530
 531 static void kvm__setup_sregs(struct kvm *self)
 532 {
 533
 534         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &self->sregs) < 0)
 535                 die_perror("KVM_GET_SREGS failed");
 536
 537         self->sregs.cs.selector = self->boot_selector;
 538         self->sregs.cs.base     = selector_to_base(self->boot_selector);
 539         self->sregs.ss.selector = self->boot_selector;
 540         self->sregs.ss.base     = selector_to_base(self->boot_selector);
 541         self->sregs.ds.selector = self->boot_selector;
 542         self->sregs.ds.base     = selector_to_base(self->boot_selector);
 543         self->sregs.es.selector = self->boot_selector;
 544         self->sregs.es.base     = selector_to_base(self->boot_selector);
 545         self->sregs.fs.selector = self->boot_selector;
 546         self->sregs.fs.base     = selector_to_base(self->boot_selector);
 547         self->sregs.gs.selector = self->boot_selector;
 548         self->sregs.gs.base     = selector_to_base(self->boot_selector);
 549
 550         if (ioctl(self->vcpu_fd, KVM_SET_SREGS, &self->sregs) < 0)
 551                 die_perror("KVM_SET_SREGS failed");
 552 }
 553
 554 /**
 555  * kvm__reset_vcpu - reset virtual CPU to a known state
 556  */
 557 void kvm__reset_vcpu(struct kvm *self)
 558 {
 559         kvm__setup_sregs(self);
 560         kvm__setup_regs(self);
 561         kvm__setup_fpu(self);
 562         kvm__setup_msrs(self);
 563 }
 564
 565 /**
 566  * kvm__setup_bios - inject BIOS into guest system memory
 567  * @self - guest system descriptor
 568  *
 569  * This function is a main routine where we poke guest memory
 570  * and install BIOS there.
 571  */
 572 void kvm__setup_bios(struct kvm *self)
 573 {
 574         /* standart minimal configuration */
 575         setup_bios(self);
 576
 577         /* FIXME: SMP, ACPI and friends here */
 578 }
 579
 580 #define TIMER_INTERVAL_NS 1000000       /* 1 msec */
 581
 582 static void alarm_handler(int sig)
 583 {
 584 }
 585
 586 /*
 587  * This function sets up a timer that's used to inject interrupts from the
 588  * userspace hypervisor into the guest at periodical intervals. Please note
 589  * that clock interrupt, for example, is not handled here.
 590  */
 591 void kvm__start_timer(struct kvm *self)
 592 {
 593         struct itimerspec its;
 594         struct sigaction sa;
 595         struct sigevent sev;
 596
 597         sigfillset(&sa.sa_mask);
 598         sa.sa_flags                     = 0;
 599         sa.sa_handler                   = alarm_handler;
 600
 601         sigaction(SIGALRM, &sa, NULL);
 602
 603         memset(&sev, 0, sizeof(struct sigevent));
 604         sev.sigev_value.sival_int       = 0;
 605         sev.sigev_notify                = SIGEV_SIGNAL;
 606         sev.sigev_signo                 = SIGALRM;
 607
 608         if (timer_create(CLOCK_REALTIME, &sev, &self->timerid) < 0)
 609                 die("timer_create()");
 610
 611         its.it_value.tv_sec             = TIMER_INTERVAL_NS / 1000000000;
 612         its.it_value.tv_nsec            = TIMER_INTERVAL_NS % 1000000000;
 613         its.it_interval.tv_sec          = its.it_value.tv_sec;
 614         its.it_interval.tv_nsec         = its.it_value.tv_nsec;
 615
 616         if (timer_settime(self->timerid, 0, &its, NULL) < 0)
 617                 die("timer_settime()");
 618 }
 619
 620 void kvm__stop_timer(struct kvm *self)
 621 {
 622         if (self->timerid)
 623                 if (timer_delete(self->timerid) < 0)
 624                         die("timer_delete()");
 625
 626         self->timerid = 0;
 627 }
 628
 629 void kvm__run(struct kvm *self)
 630 {
 631         int err;
 632
 633         err = ioctl(self->vcpu_fd, KVM_RUN, 0);
 634         if (err && (errno != EINTR && errno != EAGAIN))
 635                 die_perror("KVM_RUN failed");
 636 }
 637
 638 void kvm__irq_line(struct kvm *self, int irq, int level)
 639 {
 640         struct kvm_irq_level irq_level;
 641
 642         irq_level       = (struct kvm_irq_level) {
 643                 {
 644                         .irq            = irq,
 645                 },
 646                 .level          = level,
 647         };
 648
 649         if (ioctl(self->vm_fd, KVM_IRQ_LINE, &irq_level) < 0)
 650                 die_perror("KVM_IRQ_LINE failed");
 651 }
 652
 653 static void print_dtable(const char *name, struct kvm_dtable *dtable)
 654 {
 655         printf(" %s                 %016" PRIx64 "  %08" PRIx16 "\n",
 656                 name, (uint64_t) dtable->base, (uint16_t) dtable->limit);
 657 }
 658
 659 static void print_segment(const char *name, struct kvm_segment *seg)
 660 {
 661         printf(" %s       %04" PRIx16 "      %016" PRIx64 "  %08" PRIx32 "  %02" PRIx8 "    %x %x   %x  %x %x %x %x\n",
 662                 name, (uint16_t) seg->selector, (uint64_t) seg->base, (uint32_t) seg->limit,
 663                 (uint8_t) seg->type, seg->present, seg->dpl, seg->db, seg->s, seg->l, seg->g, seg->avl);
 664 }
 665
 666 void kvm__show_registers(struct kvm *self)
 667 {
 668         unsigned long cr0, cr2, cr3;
 669         unsigned long cr4, cr8;
 670         unsigned long rax, rbx, rcx;
 671         unsigned long rdx, rsi, rdi;
 672         unsigned long rbp,  r8,  r9;
 673         unsigned long r10, r11, r12;
 674         unsigned long r13, r14, r15;
 675         unsigned long rip, rsp;
 676         struct kvm_sregs sregs;
 677         unsigned long rflags;
 678         struct kvm_regs regs;
 679         int i;
 680
 681         if (ioctl(self->vcpu_fd, KVM_GET_REGS, &regs) < 0)
 682                 die("KVM_GET_REGS failed");
 683
 684         rflags = regs.rflags;
 685
 686         rip = regs.rip; rsp = regs.rsp;
 687         rax = regs.rax; rbx = regs.rbx; rcx = regs.rcx;
 688         rdx = regs.rdx; rsi = regs.rsi; rdi = regs.rdi;
 689         rbp = regs.rbp; r8  = regs.r8;  r9  = regs.r9;
 690         r10 = regs.r10; r11 = regs.r11; r12 = regs.r12;
 691         r13 = regs.r13; r14 = regs.r14; r15 = regs.r15;
 692
 693         printf("Registers:\n");
 694         printf(" rip: %016lx   rsp: %016lx flags: %016lx\n", rip, rsp, rflags);
 695         printf(" rax: %016lx   rbx: %016lx   rcx: %016lx\n", rax, rbx, rcx);
 696         printf(" rdx: %016lx   rsi: %016lx   rdi: %016lx\n", rdx, rsi, rdi);
 697         printf(" rbp: %016lx   r8:  %016lx   r9:  %016lx\n", rbp, r8,  r9);
 698         printf(" r10: %016lx   r11: %016lx   r12: %016lx\n", r10, r11, r12);
 699         printf(" r13: %016lx   r14: %016lx   r15: %016lx\n", r13, r14, r15);
 700
 701         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &sregs) < 0)
 702                 die("KVM_GET_REGS failed");
 703
 704         cr0 = sregs.cr0; cr2 = sregs.cr2; cr3 = sregs.cr3;
 705         cr4 = sregs.cr4; cr8 = sregs.cr8;
 706
 707         printf(" cr0: %016lx   cr2: %016lx   cr3: %016lx\n", cr0, cr2, cr3);
 708         printf(" cr4: %016lx   cr8: %016lx\n", cr4, cr8);
 709         printf("Segment registers:\n");
 710         printf(" register  selector  base              limit     type  p dpl db s l g avl\n");
 711         print_segment("cs ", &sregs.cs);
 712         print_segment("ss ", &sregs.ss);
 713         print_segment("ds ", &sregs.ds);
 714         print_segment("es ", &sregs.es);
 715         print_segment("fs ", &sregs.fs);
 716         print_segment("gs ", &sregs.gs);
 717         print_segment("tr ", &sregs.tr);
 718         print_segment("ldt", &sregs.ldt);
 719         print_dtable("gdt", &sregs.gdt);
 720         print_dtable("idt", &sregs.idt);
 721         printf(" [ efer: %016" PRIx64 "  apic base: %016" PRIx64 "  nmi: %s ]\n",
 722                 (uint64_t) sregs.efer, (uint64_t) sregs.apic_base,
 723                 (self->nmi_disabled ? "disabled" : "enabled"));
 724         printf("Interrupt bitmap:\n");
 725         printf(" ");
 726         for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++)
 727                 printf("%016" PRIx64 " ", (uint64_t) sregs.interrupt_bitmap[i]);
 728         printf("\n");
 729 }
 730
 731 void kvm__show_code(struct kvm *self)
 732 {
 733         unsigned int code_bytes = 64;
 734         unsigned int code_prologue = code_bytes * 43 / 64;
 735         unsigned int code_len = code_bytes;
 736         unsigned char c;
 737         unsigned int i;
 738         uint8_t *ip;
 739
 740         if (ioctl(self->vcpu_fd, KVM_GET_REGS, &self->regs) < 0)
 741                 die("KVM_GET_REGS failed");
 742
 743         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &self->sregs) < 0)
 744                 die("KVM_GET_SREGS failed");
 745
 746         ip = guest_flat_to_host(self, ip_to_flat(self, self->regs.rip) - code_prologue);
 747
 748         printf("Code: ");
 749
 750         for (i = 0; i < code_len; i++, ip++) {
 751                 if (!host_ptr_in_ram(self, ip))
 752                         break;
 753
 754                 c = *ip;
 755
 756                 if (ip == guest_flat_to_host(self, ip_to_flat(self, self->regs.rip)))
 757                         printf("<%02x> ", c);
 758                 else
 759                         printf("%02x ", c);
 760         }
 761
 762         printf("\n");
 763
 764         printf("Stack:\n");
 765         kvm__dump_mem(self, self->regs.rsp, 32);
 766 }
 767
 768 void kvm__show_page_tables(struct kvm *self)
 769 {
 770         uint64_t *pte1;
 771         uint64_t *pte2;
 772         uint64_t *pte3;
 773         uint64_t *pte4;
 774
 775         if (!is_in_protected_mode(self))
 776                 return;
 777
 778         if (ioctl(self->vcpu_fd, KVM_GET_SREGS, &self->sregs) < 0)
 779                 die("KVM_GET_SREGS failed");
 780
 781         pte4    = guest_flat_to_host(self, self->sregs.cr3);
 782         if (!host_ptr_in_ram(self, pte4))
 783                 return;
 784
 785         pte3    = guest_flat_to_host(self, (*pte4 & ~0xfff));
 786         if (!host_ptr_in_ram(self, pte3))
 787                 return;
 788
 789         pte2    = guest_flat_to_host(self, (*pte3 & ~0xfff));
 790         if (!host_ptr_in_ram(self, pte2))
 791                 return;
 792
 793         pte1    = guest_flat_to_host(self, (*pte2 & ~0xfff));
 794         if (!host_ptr_in_ram(self, pte1))
 795                 return;
 796
 797         printf("Page Tables:\n");
 798         if (*pte2 & (1 << 7))
 799                 printf(" pte4: %016" PRIx64 "   pte3: %016" PRIx64
 800                         "   pte2: %016" PRIx64 "\n",
 801                         *pte4, *pte3, *pte2);
 802         else
 803                 printf(" pte4: %016" PRIx64 "   pte3: %016" PRIx64 "   pte2: %016"
 804                         PRIx64 "   pte1: %016" PRIx64 "\n",
 805                         *pte4, *pte3, *pte2, *pte1);
 806 }
 807
 808 void kvm__dump_mem(struct kvm *self, unsigned long addr, unsigned long size)
 809 {
 810         unsigned char *p;
 811         unsigned long n;
 812
 813         size &= ~7; /* mod 8 */
 814         if (!size)
 815                 return;
 816
 817         p = guest_flat_to_host(self, addr);
 818
 819         for (n = 0; n < size; n+=8) {
 820                 if (!host_ptr_in_ram(self, p + n))
 821                         break;
 822
 823                 printf("  0x%08lx: %02x %02x %02x %02x  %02x %02x %02x %02x\n",
 824                         addr + n, p[n + 0], p[n + 1], p[n + 2], p[n + 3],
 825                                   p[n + 4], p[n + 5], p[n + 6], p[n + 7]);
 826         }
 827 }