1 #include "kvm/kvm-cpu.h" 2 3 #include "kvm/symbol.h" 4 #include "kvm/util.h" 5 #include "kvm/kvm.h" 6 7 #include <asm/msr-index.h> 8 9 #include <sys/ioctl.h> 10 #include <sys/mman.h> 11 #include <signal.h> 12 #include <stdlib.h> 13 #include <string.h> 14 #include <errno.h> 15 #include <stdio.h> 16 17 #define PAGE_SIZE (sysconf(_SC_PAGE_SIZE)) 18 19 extern __thread struct kvm_cpu *current_kvm_cpu; 20 21 static inline bool is_in_protected_mode(struct kvm_cpu *vcpu) 22 { 23 return vcpu->sregs.cr0 & 0x01; 24 } 25 26 static inline u64 ip_to_flat(struct kvm_cpu *vcpu, u64 ip) 27 { 28 u64 cs; 29 30 /* 31 * NOTE! We should take code segment base address into account here. 32 * Luckily it's usually zero because Linux uses flat memory model. 33 */ 34 if (is_in_protected_mode(vcpu)) 35 return ip; 36 37 cs = vcpu->sregs.cs.selector; 38 39 return ip + (cs << 4); 40 } 41 42 static inline u32 selector_to_base(u16 selector) 43 { 44 /* 45 * KVM on Intel requires 'base' to be 'selector * 16' in real mode. 46 */ 47 return (u32)selector * 16; 48 } 49 50 static struct kvm_cpu *kvm_cpu__new(struct kvm *kvm) 51 { 52 struct kvm_cpu *vcpu; 53 54 vcpu = calloc(1, sizeof *vcpu); 55 if (!vcpu) 56 return NULL; 57 58 vcpu->kvm = kvm; 59 60 return vcpu; 61 } 62 63 void kvm_cpu__delete(struct kvm_cpu *vcpu) 64 { 65 if (vcpu->msrs) 66 free(vcpu->msrs); 67 68 free(vcpu); 69 } 70 71 struct kvm_cpu *kvm_cpu__init(struct kvm *kvm, unsigned long cpu_id) 72 { 73 struct kvm_cpu *vcpu; 74 int mmap_size; 75 int coalesced_offset; 76 77 vcpu = kvm_cpu__new(kvm); 78 if (!vcpu) 79 return NULL; 80 81 vcpu->cpu_id = cpu_id; 82 83 vcpu->vcpu_fd = ioctl(vcpu->kvm->vm_fd, KVM_CREATE_VCPU, cpu_id); 84 if (vcpu->vcpu_fd < 0) 85 die_perror("KVM_CREATE_VCPU ioctl"); 86 87 mmap_size = ioctl(vcpu->kvm->sys_fd, KVM_GET_VCPU_MMAP_SIZE, 0); 88 if (mmap_size < 0) 89 die_perror("KVM_GET_VCPU_MMAP_SIZE ioctl"); 90 91 vcpu->kvm_run = mmap(NULL, mmap_size, PROT_RW, MAP_SHARED, vcpu->vcpu_fd, 0); 92 if (vcpu->kvm_run == MAP_FAILED) 93 die("unable to mmap vcpu fd"); 94 95 coalesced_offset = ioctl(kvm->sys_fd, KVM_CHECK_EXTENSION, KVM_CAP_COALESCED_MMIO); 96 if (coalesced_offset) 97 vcpu->ring = (void *)vcpu->kvm_run + (coalesced_offset * PAGE_SIZE); 98 99 vcpu->is_running = true; 100 101 return vcpu; 102 } 103 104 void kvm_cpu__enable_singlestep(struct kvm_cpu *vcpu) 105 { 106 struct kvm_guest_debug debug = { 107 .control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP, 108 }; 109 110 if (ioctl(vcpu->vcpu_fd, KVM_SET_GUEST_DEBUG, &debug) < 0) 111 pr_warning("KVM_SET_GUEST_DEBUG failed"); 112 } 113 114 static struct kvm_msrs *kvm_msrs__new(size_t nmsrs) 115 { 116 struct kvm_msrs *vcpu = calloc(1, sizeof(*vcpu) + (sizeof(struct kvm_msr_entry) * nmsrs)); 117 118 if (!vcpu) 119 die("out of memory"); 120 121 return vcpu; 122 } 123 124 #define KVM_MSR_ENTRY(_index, _data) \ 125 (struct kvm_msr_entry) { .index = _index, .data = _data } 126 127 static void kvm_cpu__setup_msrs(struct kvm_cpu *vcpu) 128 { 129 unsigned long ndx = 0; 130 131 vcpu->msrs = kvm_msrs__new(100); 132 133 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_CS, 0x0); 134 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_ESP, 0x0); 135 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_EIP, 0x0); 136 #ifdef CONFIG_X86_64 137 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_STAR, 0x0); 138 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_CSTAR, 0x0); 139 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_KERNEL_GS_BASE, 0x0); 140 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_SYSCALL_MASK, 0x0); 141 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_LSTAR, 0x0); 142 #endif 143 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_TSC, 0x0); 144 145 vcpu->msrs->nmsrs = ndx; 146 147 if (ioctl(vcpu->vcpu_fd, KVM_SET_MSRS, vcpu->msrs) < 0) 148 die_perror("KVM_SET_MSRS failed"); 149 } 150 151 static void kvm_cpu__setup_fpu(struct kvm_cpu *vcpu) 152 { 153 vcpu->fpu = (struct kvm_fpu) { 154 .fcw = 0x37f, 155 .mxcsr = 0x1f80, 156 }; 157 158 if (ioctl(vcpu->vcpu_fd, KVM_SET_FPU, &vcpu->fpu) < 0) 159 die_perror("KVM_SET_FPU failed"); 160 } 161 162 static void kvm_cpu__setup_regs(struct kvm_cpu *vcpu) 163 { 164 vcpu->regs = (struct kvm_regs) { 165 /* We start the guest in 16-bit real mode */ 166 .rflags = 0x0000000000000002ULL, 167 168 .rip = vcpu->kvm->boot_ip, 169 .rsp = vcpu->kvm->boot_sp, 170 .rbp = vcpu->kvm->boot_sp, 171 }; 172 173 if (vcpu->regs.rip > USHRT_MAX) 174 die("ip 0x%llx is too high for real mode", (u64) vcpu->regs.rip); 175 176 if (ioctl(vcpu->vcpu_fd, KVM_SET_REGS, &vcpu->regs) < 0) 177 die_perror("KVM_SET_REGS failed"); 178 } 179 180 static void kvm_cpu__setup_sregs(struct kvm_cpu *vcpu) 181 { 182 183 if (ioctl(vcpu->vcpu_fd, KVM_GET_SREGS, &vcpu->sregs) < 0) 184 die_perror("KVM_GET_SREGS failed"); 185 186 vcpu->sregs.cs.selector = vcpu->kvm->boot_selector; 187 vcpu->sregs.cs.base = selector_to_base(vcpu->kvm->boot_selector); 188 vcpu->sregs.ss.selector = vcpu->kvm->boot_selector; 189 vcpu->sregs.ss.base = selector_to_base(vcpu->kvm->boot_selector); 190 vcpu->sregs.ds.selector = vcpu->kvm->boot_selector; 191 vcpu->sregs.ds.base = selector_to_base(vcpu->kvm->boot_selector); 192 vcpu->sregs.es.selector = vcpu->kvm->boot_selector; 193 vcpu->sregs.es.base = selector_to_base(vcpu->kvm->boot_selector); 194 vcpu->sregs.fs.selector = vcpu->kvm->boot_selector; 195 vcpu->sregs.fs.base = selector_to_base(vcpu->kvm->boot_selector); 196 vcpu->sregs.gs.selector = vcpu->kvm->boot_selector; 197 vcpu->sregs.gs.base = selector_to_base(vcpu->kvm->boot_selector); 198 199 if (ioctl(vcpu->vcpu_fd, KVM_SET_SREGS, &vcpu->sregs) < 0) 200 die_perror("KVM_SET_SREGS failed"); 201 } 202 203 /** 204 * kvm_cpu__reset_vcpu - reset virtual CPU to a known state 205 */ 206 void kvm_cpu__reset_vcpu(struct kvm_cpu *vcpu) 207 { 208 kvm_cpu__setup_sregs(vcpu); 209 kvm_cpu__setup_regs(vcpu); 210 kvm_cpu__setup_fpu(vcpu); 211 kvm_cpu__setup_msrs(vcpu); 212 } 213 214 static void print_dtable(const char *name, struct kvm_dtable *dtable) 215 { 216 printf(" %s %016llx %08hx\n", 217 name, (u64) dtable->base, (u16) dtable->limit); 218 } 219 220 static void print_segment(const char *name, struct kvm_segment *seg) 221 { 222 printf(" %s %04hx %016llx %08x %02hhx %x %x %x %x %x %x %x\n", 223 name, (u16) seg->selector, (u64) seg->base, (u32) seg->limit, 224 (u8) seg->type, seg->present, seg->dpl, seg->db, seg->s, seg->l, seg->g, seg->avl); 225 } 226 227 void kvm_cpu__show_registers(struct kvm_cpu *vcpu) 228 { 229 unsigned long cr0, cr2, cr3; 230 unsigned long cr4, cr8; 231 unsigned long rax, rbx, rcx; 232 unsigned long rdx, rsi, rdi; 233 unsigned long rbp, r8, r9; 234 unsigned long r10, r11, r12; 235 unsigned long r13, r14, r15; 236 unsigned long rip, rsp; 237 struct kvm_sregs sregs; 238 unsigned long rflags; 239 struct kvm_regs regs; 240 int i; 241 242 if (ioctl(vcpu->vcpu_fd, KVM_GET_REGS, ®s) < 0) 243 die("KVM_GET_REGS failed"); 244 245 rflags = regs.rflags; 246 247 rip = regs.rip; rsp = regs.rsp; 248 rax = regs.rax; rbx = regs.rbx; rcx = regs.rcx; 249 rdx = regs.rdx; rsi = regs.rsi; rdi = regs.rdi; 250 rbp = regs.rbp; r8 = regs.r8; r9 = regs.r9; 251 r10 = regs.r10; r11 = regs.r11; r12 = regs.r12; 252 r13 = regs.r13; r14 = regs.r14; r15 = regs.r15; 253 254 printf("\n Registers:\n"); 255 printf( " ----------\n"); 256 printf(" rip: %016lx rsp: %016lx flags: %016lx\n", rip, rsp, rflags); 257 printf(" rax: %016lx rbx: %016lx rcx: %016lx\n", rax, rbx, rcx); 258 printf(" rdx: %016lx rsi: %016lx rdi: %016lx\n", rdx, rsi, rdi); 259 printf(" rbp: %016lx r8: %016lx r9: %016lx\n", rbp, r8, r9); 260 printf(" r10: %016lx r11: %016lx r12: %016lx\n", r10, r11, r12); 261 printf(" r13: %016lx r14: %016lx r15: %016lx\n", r13, r14, r15); 262 263 if (ioctl(vcpu->vcpu_fd, KVM_GET_SREGS, &sregs) < 0) 264 die("KVM_GET_REGS failed"); 265 266 cr0 = sregs.cr0; cr2 = sregs.cr2; cr3 = sregs.cr3; 267 cr4 = sregs.cr4; cr8 = sregs.cr8; 268 269 printf(" cr0: %016lx cr2: %016lx cr3: %016lx\n", cr0, cr2, cr3); 270 printf(" cr4: %016lx cr8: %016lx\n", cr4, cr8); 271 printf("\n Segment registers:\n"); 272 printf( " ------------------\n"); 273 printf(" register selector base limit type p dpl db s l g avl\n"); 274 print_segment("cs ", &sregs.cs); 275 print_segment("ss ", &sregs.ss); 276 print_segment("ds ", &sregs.ds); 277 print_segment("es ", &sregs.es); 278 print_segment("fs ", &sregs.fs); 279 print_segment("gs ", &sregs.gs); 280 print_segment("tr ", &sregs.tr); 281 print_segment("ldt", &sregs.ldt); 282 print_dtable("gdt", &sregs.gdt); 283 print_dtable("idt", &sregs.idt); 284 285 printf("\n APIC:\n"); 286 printf( " -----\n"); 287 printf(" efer: %016llx apic base: %016llx nmi: %s\n", 288 (u64) sregs.efer, (u64) sregs.apic_base, 289 (vcpu->kvm->nmi_disabled ? "disabled" : "enabled")); 290 291 printf("\n Interrupt bitmap:\n"); 292 printf( " -----------------\n"); 293 for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++) 294 printf(" %016llx", (u64) sregs.interrupt_bitmap[i]); 295 printf("\n"); 296 } 297 298 #define MAX_SYM_LEN 128 299 300 void kvm_cpu__show_code(struct kvm_cpu *vcpu) 301 { 302 unsigned int code_bytes = 64; 303 unsigned int code_prologue = code_bytes * 43 / 64; 304 unsigned int code_len = code_bytes; 305 char sym[MAX_SYM_LEN]; 306 unsigned char c; 307 unsigned int i; 308 u8 *ip; 309 310 if (ioctl(vcpu->vcpu_fd, KVM_GET_REGS, &vcpu->regs) < 0) 311 die("KVM_GET_REGS failed"); 312 313 if (ioctl(vcpu->vcpu_fd, KVM_GET_SREGS, &vcpu->sregs) < 0) 314 die("KVM_GET_SREGS failed"); 315 316 ip = guest_flat_to_host(vcpu->kvm, ip_to_flat(vcpu, vcpu->regs.rip) - code_prologue); 317 318 printf("\n Code:\n"); 319 printf( " -----\n"); 320 321 symbol__lookup(vcpu->kvm, vcpu->regs.rip, sym, MAX_SYM_LEN); 322 323 printf(" rip: [<%016lx>] %s\n\n", (unsigned long) vcpu->regs.rip, sym); 324 325 for (i = 0; i < code_len; i++, ip++) { 326 if (!host_ptr_in_ram(vcpu->kvm, ip)) 327 break; 328 329 c = *ip; 330 331 if (ip == guest_flat_to_host(vcpu->kvm, ip_to_flat(vcpu, vcpu->regs.rip))) 332 printf(" <%02x>", c); 333 else 334 printf(" %02x", c); 335 } 336 337 printf("\n"); 338 339 printf("\n Stack:\n"); 340 printf( " ------\n"); 341 kvm__dump_mem(vcpu->kvm, vcpu->regs.rsp, 32); 342 } 343 344 void kvm_cpu__show_page_tables(struct kvm_cpu *vcpu) 345 { 346 u64 *pte1; 347 u64 *pte2; 348 u64 *pte3; 349 u64 *pte4; 350 351 if (!is_in_protected_mode(vcpu)) 352 return; 353 354 if (ioctl(vcpu->vcpu_fd, KVM_GET_SREGS, &vcpu->sregs) < 0) 355 die("KVM_GET_SREGS failed"); 356 357 pte4 = guest_flat_to_host(vcpu->kvm, vcpu->sregs.cr3); 358 if (!host_ptr_in_ram(vcpu->kvm, pte4)) 359 return; 360 361 pte3 = guest_flat_to_host(vcpu->kvm, (*pte4 & ~0xfff)); 362 if (!host_ptr_in_ram(vcpu->kvm, pte3)) 363 return; 364 365 pte2 = guest_flat_to_host(vcpu->kvm, (*pte3 & ~0xfff)); 366 if (!host_ptr_in_ram(vcpu->kvm, pte2)) 367 return; 368 369 pte1 = guest_flat_to_host(vcpu->kvm, (*pte2 & ~0xfff)); 370 if (!host_ptr_in_ram(vcpu->kvm, pte1)) 371 return; 372 373 printf("Page Tables:\n"); 374 if (*pte2 & (1 << 7)) 375 printf(" pte4: %016llx pte3: %016llx" 376 " pte2: %016llx\n", 377 *pte4, *pte3, *pte2); 378 else 379 printf(" pte4: %016llx pte3: %016llx pte2: %016" 380 "llx pte1: %016llx\n", 381 *pte4, *pte3, *pte2, *pte1); 382 } 383 384 void kvm_cpu__run(struct kvm_cpu *vcpu) 385 { 386 int err; 387 388 err = ioctl(vcpu->vcpu_fd, KVM_RUN, 0); 389 if (err && (errno != EINTR && errno != EAGAIN)) 390 die_perror("KVM_RUN failed"); 391 } 392 393 static void kvm_cpu_signal_handler(int signum) 394 { 395 if (signum == SIGKVMEXIT) { 396 if (current_kvm_cpu->is_running) { 397 current_kvm_cpu->is_running = false; 398 pthread_kill(pthread_self(), SIGKVMEXIT); 399 } 400 } else if (signum == SIGKVMPAUSE) { 401 current_kvm_cpu->paused = 1; 402 } 403 } 404 405 static void kvm_cpu__handle_coalesced_mmio(struct kvm_cpu *cpu) 406 { 407 if (cpu->ring) { 408 while (cpu->ring->first != cpu->ring->last) { 409 struct kvm_coalesced_mmio *m; 410 m = &cpu->ring->coalesced_mmio[cpu->ring->first]; 411 kvm__emulate_mmio(cpu->kvm, 412 m->phys_addr, 413 m->data, 414 m->len, 415 1); 416 cpu->ring->first = (cpu->ring->first + 1) % KVM_COALESCED_MMIO_MAX; 417 } 418 } 419 } 420 421 int kvm_cpu__start(struct kvm_cpu *cpu) 422 { 423 sigset_t sigset; 424 425 sigemptyset(&sigset); 426 sigaddset(&sigset, SIGALRM); 427 428 pthread_sigmask(SIG_BLOCK, &sigset, NULL); 429 430 signal(SIGKVMEXIT, kvm_cpu_signal_handler); 431 signal(SIGKVMPAUSE, kvm_cpu_signal_handler); 432 433 kvm_cpu__setup_cpuid(cpu); 434 kvm_cpu__reset_vcpu(cpu); 435 436 for (;;) { 437 if (cpu->paused) { 438 kvm__notify_paused(); 439 cpu->paused = 0; 440 } 441 442 kvm_cpu__run(cpu); 443 444 switch (cpu->kvm_run->exit_reason) { 445 case KVM_EXIT_UNKNOWN: 446 break; 447 case KVM_EXIT_DEBUG: 448 kvm_cpu__show_registers(cpu); 449 kvm_cpu__show_code(cpu); 450 break; 451 case KVM_EXIT_IO: { 452 bool ret; 453 454 ret = kvm__emulate_io(cpu->kvm, 455 cpu->kvm_run->io.port, 456 (u8 *)cpu->kvm_run + 457 cpu->kvm_run->io.data_offset, 458 cpu->kvm_run->io.direction, 459 cpu->kvm_run->io.size, 460 cpu->kvm_run->io.count); 461 462 if (!ret) 463 goto panic_kvm; 464 break; 465 } 466 case KVM_EXIT_MMIO: { 467 bool ret; 468 469 ret = kvm__emulate_mmio(cpu->kvm, 470 cpu->kvm_run->mmio.phys_addr, 471 cpu->kvm_run->mmio.data, 472 cpu->kvm_run->mmio.len, 473 cpu->kvm_run->mmio.is_write); 474 475 if (!ret) 476 goto panic_kvm; 477 break; 478 } 479 case KVM_EXIT_INTR: 480 if (cpu->is_running) 481 break; 482 goto exit_kvm; 483 case KVM_EXIT_SHUTDOWN: 484 goto exit_kvm; 485 default: 486 goto panic_kvm; 487 } 488 kvm_cpu__handle_coalesced_mmio(cpu); 489 } 490 491 exit_kvm: 492 return 0; 493 494 panic_kvm: 495 return 1; 496 } 497