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 struct kvm_cpu *kvm_cpus[KVM_NR_CPUS]; 20 extern __thread struct kvm_cpu *current_kvm_cpu; 21 22 static inline bool is_in_protected_mode(struct kvm_cpu *vcpu) 23 { 24 return vcpu->sregs.cr0 & 0x01; 25 } 26 27 static inline u64 ip_to_flat(struct kvm_cpu *vcpu, u64 ip) 28 { 29 u64 cs; 30 31 /* 32 * NOTE! We should take code segment base address into account here. 33 * Luckily it's usually zero because Linux uses flat memory model. 34 */ 35 if (is_in_protected_mode(vcpu)) 36 return ip; 37 38 cs = vcpu->sregs.cs.selector; 39 40 return ip + (cs << 4); 41 } 42 43 static inline u32 selector_to_base(u16 selector) 44 { 45 /* 46 * KVM on Intel requires 'base' to be 'selector * 16' in real mode. 47 */ 48 return (u32)selector * 16; 49 } 50 51 static struct kvm_cpu *kvm_cpu__new(struct kvm *kvm) 52 { 53 struct kvm_cpu *vcpu; 54 55 vcpu = calloc(1, sizeof *vcpu); 56 if (!vcpu) 57 return NULL; 58 59 vcpu->kvm = kvm; 60 61 return vcpu; 62 } 63 64 void kvm_cpu__delete(struct kvm_cpu *vcpu) 65 { 66 if (vcpu->msrs) 67 free(vcpu->msrs); 68 69 free(vcpu); 70 } 71 72 struct kvm_cpu *kvm_cpu__init(struct kvm *kvm, unsigned long cpu_id) 73 { 74 struct kvm_cpu *vcpu; 75 int mmap_size; 76 int coalesced_offset; 77 78 vcpu = kvm_cpu__new(kvm); 79 if (!vcpu) 80 return NULL; 81 82 vcpu->cpu_id = cpu_id; 83 84 vcpu->vcpu_fd = ioctl(vcpu->kvm->vm_fd, KVM_CREATE_VCPU, cpu_id); 85 if (vcpu->vcpu_fd < 0) 86 die_perror("KVM_CREATE_VCPU ioctl"); 87 88 mmap_size = ioctl(vcpu->kvm->sys_fd, KVM_GET_VCPU_MMAP_SIZE, 0); 89 if (mmap_size < 0) 90 die_perror("KVM_GET_VCPU_MMAP_SIZE ioctl"); 91 92 vcpu->kvm_run = mmap(NULL, mmap_size, PROT_RW, MAP_SHARED, vcpu->vcpu_fd, 0); 93 if (vcpu->kvm_run == MAP_FAILED) 94 die("unable to mmap vcpu fd"); 95 96 coalesced_offset = ioctl(kvm->sys_fd, KVM_CHECK_EXTENSION, KVM_CAP_COALESCED_MMIO); 97 if (coalesced_offset) 98 vcpu->ring = (void *)vcpu->kvm_run + (coalesced_offset * PAGE_SIZE); 99 100 vcpu->is_running = true; 101 102 return vcpu; 103 } 104 105 void kvm_cpu__enable_singlestep(struct kvm_cpu *vcpu) 106 { 107 struct kvm_guest_debug debug = { 108 .control = KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP, 109 }; 110 111 if (ioctl(vcpu->vcpu_fd, KVM_SET_GUEST_DEBUG, &debug) < 0) 112 pr_warning("KVM_SET_GUEST_DEBUG failed"); 113 } 114 115 static struct kvm_msrs *kvm_msrs__new(size_t nmsrs) 116 { 117 struct kvm_msrs *vcpu = calloc(1, sizeof(*vcpu) + (sizeof(struct kvm_msr_entry) * nmsrs)); 118 119 if (!vcpu) 120 die("out of memory"); 121 122 return vcpu; 123 } 124 125 #define KVM_MSR_ENTRY(_index, _data) \ 126 (struct kvm_msr_entry) { .index = _index, .data = _data } 127 128 static void kvm_cpu__setup_msrs(struct kvm_cpu *vcpu) 129 { 130 unsigned long ndx = 0; 131 132 vcpu->msrs = kvm_msrs__new(100); 133 134 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_CS, 0x0); 135 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_ESP, 0x0); 136 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_SYSENTER_EIP, 0x0); 137 #ifdef CONFIG_X86_64 138 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_STAR, 0x0); 139 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_CSTAR, 0x0); 140 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_KERNEL_GS_BASE, 0x0); 141 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_SYSCALL_MASK, 0x0); 142 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_LSTAR, 0x0); 143 #endif 144 vcpu->msrs->entries[ndx++] = KVM_MSR_ENTRY(MSR_IA32_TSC, 0x0); 145 146 vcpu->msrs->nmsrs = ndx; 147 148 if (ioctl(vcpu->vcpu_fd, KVM_SET_MSRS, vcpu->msrs) < 0) 149 die_perror("KVM_SET_MSRS failed"); 150 } 151 152 static void kvm_cpu__setup_fpu(struct kvm_cpu *vcpu) 153 { 154 vcpu->fpu = (struct kvm_fpu) { 155 .fcw = 0x37f, 156 .mxcsr = 0x1f80, 157 }; 158 159 if (ioctl(vcpu->vcpu_fd, KVM_SET_FPU, &vcpu->fpu) < 0) 160 die_perror("KVM_SET_FPU failed"); 161 } 162 163 static void kvm_cpu__setup_regs(struct kvm_cpu *vcpu) 164 { 165 vcpu->regs = (struct kvm_regs) { 166 /* We start the guest in 16-bit real mode */ 167 .rflags = 0x0000000000000002ULL, 168 169 .rip = vcpu->kvm->boot_ip, 170 .rsp = vcpu->kvm->boot_sp, 171 .rbp = vcpu->kvm->boot_sp, 172 }; 173 174 if (vcpu->regs.rip > USHRT_MAX) 175 die("ip 0x%llx is too high for real mode", (u64) vcpu->regs.rip); 176 177 if (ioctl(vcpu->vcpu_fd, KVM_SET_REGS, &vcpu->regs) < 0) 178 die_perror("KVM_SET_REGS failed"); 179 } 180 181 static void kvm_cpu__setup_sregs(struct kvm_cpu *vcpu) 182 { 183 184 if (ioctl(vcpu->vcpu_fd, KVM_GET_SREGS, &vcpu->sregs) < 0) 185 die_perror("KVM_GET_SREGS failed"); 186 187 vcpu->sregs.cs.selector = vcpu->kvm->boot_selector; 188 vcpu->sregs.cs.base = selector_to_base(vcpu->kvm->boot_selector); 189 vcpu->sregs.ss.selector = vcpu->kvm->boot_selector; 190 vcpu->sregs.ss.base = selector_to_base(vcpu->kvm->boot_selector); 191 vcpu->sregs.ds.selector = vcpu->kvm->boot_selector; 192 vcpu->sregs.ds.base = selector_to_base(vcpu->kvm->boot_selector); 193 vcpu->sregs.es.selector = vcpu->kvm->boot_selector; 194 vcpu->sregs.es.base = selector_to_base(vcpu->kvm->boot_selector); 195 vcpu->sregs.fs.selector = vcpu->kvm->boot_selector; 196 vcpu->sregs.fs.base = selector_to_base(vcpu->kvm->boot_selector); 197 vcpu->sregs.gs.selector = vcpu->kvm->boot_selector; 198 vcpu->sregs.gs.base = selector_to_base(vcpu->kvm->boot_selector); 199 200 if (ioctl(vcpu->vcpu_fd, KVM_SET_SREGS, &vcpu->sregs) < 0) 201 die_perror("KVM_SET_SREGS failed"); 202 } 203 204 /** 205 * kvm_cpu__reset_vcpu - reset virtual CPU to a known state 206 */ 207 void kvm_cpu__reset_vcpu(struct kvm_cpu *vcpu) 208 { 209 kvm_cpu__setup_sregs(vcpu); 210 kvm_cpu__setup_regs(vcpu); 211 kvm_cpu__setup_fpu(vcpu); 212 kvm_cpu__setup_msrs(vcpu); 213 } 214 215 static void print_dtable(const char *name, struct kvm_dtable *dtable) 216 { 217 printf(" %s %016llx %08hx\n", 218 name, (u64) dtable->base, (u16) dtable->limit); 219 } 220 221 static void print_segment(const char *name, struct kvm_segment *seg) 222 { 223 printf(" %s %04hx %016llx %08x %02hhx %x %x %x %x %x %x %x\n", 224 name, (u16) seg->selector, (u64) seg->base, (u32) seg->limit, 225 (u8) seg->type, seg->present, seg->dpl, seg->db, seg->s, seg->l, seg->g, seg->avl); 226 } 227 228 void kvm_cpu__show_registers(struct kvm_cpu *vcpu) 229 { 230 unsigned long cr0, cr2, cr3; 231 unsigned long cr4, cr8; 232 unsigned long rax, rbx, rcx; 233 unsigned long rdx, rsi, rdi; 234 unsigned long rbp, r8, r9; 235 unsigned long r10, r11, r12; 236 unsigned long r13, r14, r15; 237 unsigned long rip, rsp; 238 struct kvm_sregs sregs; 239 unsigned long rflags; 240 struct kvm_regs regs; 241 int i; 242 243 if (ioctl(vcpu->vcpu_fd, KVM_GET_REGS, ®s) < 0) 244 die("KVM_GET_REGS failed"); 245 246 rflags = regs.rflags; 247 248 rip = regs.rip; rsp = regs.rsp; 249 rax = regs.rax; rbx = regs.rbx; rcx = regs.rcx; 250 rdx = regs.rdx; rsi = regs.rsi; rdi = regs.rdi; 251 rbp = regs.rbp; r8 = regs.r8; r9 = regs.r9; 252 r10 = regs.r10; r11 = regs.r11; r12 = regs.r12; 253 r13 = regs.r13; r14 = regs.r14; r15 = regs.r15; 254 255 printf("\n Registers:\n"); 256 printf( " ----------\n"); 257 printf(" rip: %016lx rsp: %016lx flags: %016lx\n", rip, rsp, rflags); 258 printf(" rax: %016lx rbx: %016lx rcx: %016lx\n", rax, rbx, rcx); 259 printf(" rdx: %016lx rsi: %016lx rdi: %016lx\n", rdx, rsi, rdi); 260 printf(" rbp: %016lx r8: %016lx r9: %016lx\n", rbp, r8, r9); 261 printf(" r10: %016lx r11: %016lx r12: %016lx\n", r10, r11, r12); 262 printf(" r13: %016lx r14: %016lx r15: %016lx\n", r13, r14, r15); 263 264 if (ioctl(vcpu->vcpu_fd, KVM_GET_SREGS, &sregs) < 0) 265 die("KVM_GET_REGS failed"); 266 267 cr0 = sregs.cr0; cr2 = sregs.cr2; cr3 = sregs.cr3; 268 cr4 = sregs.cr4; cr8 = sregs.cr8; 269 270 printf(" cr0: %016lx cr2: %016lx cr3: %016lx\n", cr0, cr2, cr3); 271 printf(" cr4: %016lx cr8: %016lx\n", cr4, cr8); 272 printf("\n Segment registers:\n"); 273 printf( " ------------------\n"); 274 printf(" register selector base limit type p dpl db s l g avl\n"); 275 print_segment("cs ", &sregs.cs); 276 print_segment("ss ", &sregs.ss); 277 print_segment("ds ", &sregs.ds); 278 print_segment("es ", &sregs.es); 279 print_segment("fs ", &sregs.fs); 280 print_segment("gs ", &sregs.gs); 281 print_segment("tr ", &sregs.tr); 282 print_segment("ldt", &sregs.ldt); 283 print_dtable("gdt", &sregs.gdt); 284 print_dtable("idt", &sregs.idt); 285 286 printf("\n APIC:\n"); 287 printf( " -----\n"); 288 printf(" efer: %016llx apic base: %016llx nmi: %s\n", 289 (u64) sregs.efer, (u64) sregs.apic_base, 290 (vcpu->kvm->nmi_disabled ? "disabled" : "enabled")); 291 292 printf("\n Interrupt bitmap:\n"); 293 printf( " -----------------\n"); 294 for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++) 295 printf(" %016llx", (u64) sregs.interrupt_bitmap[i]); 296 printf("\n"); 297 } 298 299 #define MAX_SYM_LEN 128 300 301 void kvm_cpu__show_code(struct kvm_cpu *vcpu) 302 { 303 unsigned int code_bytes = 64; 304 unsigned int code_prologue = code_bytes * 43 / 64; 305 unsigned int code_len = code_bytes; 306 char sym[MAX_SYM_LEN]; 307 unsigned char c; 308 unsigned int i; 309 u8 *ip; 310 311 if (ioctl(vcpu->vcpu_fd, KVM_GET_REGS, &vcpu->regs) < 0) 312 die("KVM_GET_REGS failed"); 313 314 if (ioctl(vcpu->vcpu_fd, KVM_GET_SREGS, &vcpu->sregs) < 0) 315 die("KVM_GET_SREGS failed"); 316 317 ip = guest_flat_to_host(vcpu->kvm, ip_to_flat(vcpu, vcpu->regs.rip) - code_prologue); 318 319 printf("\n Code:\n"); 320 printf( " -----\n"); 321 322 symbol__lookup(vcpu->kvm, vcpu->regs.rip, sym, MAX_SYM_LEN); 323 324 printf(" rip: [<%016lx>] %s\n\n", (unsigned long) vcpu->regs.rip, sym); 325 326 for (i = 0; i < code_len; i++, ip++) { 327 if (!host_ptr_in_ram(vcpu->kvm, ip)) 328 break; 329 330 c = *ip; 331 332 if (ip == guest_flat_to_host(vcpu->kvm, ip_to_flat(vcpu, vcpu->regs.rip))) 333 printf(" <%02x>", c); 334 else 335 printf(" %02x", c); 336 } 337 338 printf("\n"); 339 340 printf("\n Stack:\n"); 341 printf( " ------\n"); 342 kvm__dump_mem(vcpu->kvm, vcpu->regs.rsp, 32); 343 } 344 345 void kvm_cpu__show_page_tables(struct kvm_cpu *vcpu) 346 { 347 u64 *pte1; 348 u64 *pte2; 349 u64 *pte3; 350 u64 *pte4; 351 352 if (!is_in_protected_mode(vcpu)) 353 return; 354 355 if (ioctl(vcpu->vcpu_fd, KVM_GET_SREGS, &vcpu->sregs) < 0) 356 die("KVM_GET_SREGS failed"); 357 358 pte4 = guest_flat_to_host(vcpu->kvm, vcpu->sregs.cr3); 359 if (!host_ptr_in_ram(vcpu->kvm, pte4)) 360 return; 361 362 pte3 = guest_flat_to_host(vcpu->kvm, (*pte4 & ~0xfff)); 363 if (!host_ptr_in_ram(vcpu->kvm, pte3)) 364 return; 365 366 pte2 = guest_flat_to_host(vcpu->kvm, (*pte3 & ~0xfff)); 367 if (!host_ptr_in_ram(vcpu->kvm, pte2)) 368 return; 369 370 pte1 = guest_flat_to_host(vcpu->kvm, (*pte2 & ~0xfff)); 371 if (!host_ptr_in_ram(vcpu->kvm, pte1)) 372 return; 373 374 printf("Page Tables:\n"); 375 if (*pte2 & (1 << 7)) 376 printf(" pte4: %016llx pte3: %016llx" 377 " pte2: %016llx\n", 378 *pte4, *pte3, *pte2); 379 else 380 printf(" pte4: %016llx pte3: %016llx pte2: %016" 381 "llx pte1: %016llx\n", 382 *pte4, *pte3, *pte2, *pte1); 383 } 384 385 void kvm_cpu__run(struct kvm_cpu *vcpu) 386 { 387 int err; 388 389 err = ioctl(vcpu->vcpu_fd, KVM_RUN, 0); 390 if (err && (errno != EINTR && errno != EAGAIN)) 391 die_perror("KVM_RUN failed"); 392 } 393 394 static void kvm_cpu_signal_handler(int signum) 395 { 396 if (signum == SIGKVMEXIT) { 397 if (current_kvm_cpu && current_kvm_cpu->is_running) { 398 current_kvm_cpu->is_running = false; 399 pthread_kill(pthread_self(), SIGKVMEXIT); 400 } 401 } else if (signum == SIGKVMPAUSE) { 402 current_kvm_cpu->paused = 1; 403 } 404 } 405 406 static void kvm_cpu__handle_coalesced_mmio(struct kvm_cpu *cpu) 407 { 408 if (cpu->ring) { 409 while (cpu->ring->first != cpu->ring->last) { 410 struct kvm_coalesced_mmio *m; 411 m = &cpu->ring->coalesced_mmio[cpu->ring->first]; 412 kvm__emulate_mmio(cpu->kvm, 413 m->phys_addr, 414 m->data, 415 m->len, 416 1); 417 cpu->ring->first = (cpu->ring->first + 1) % KVM_COALESCED_MMIO_MAX; 418 } 419 } 420 } 421 422 void kvm_cpu__reboot(void) 423 { 424 pthread_kill(kvm_cpus[0]->thread, SIGKVMEXIT); 425 } 426 427 int kvm_cpu__start(struct kvm_cpu *cpu) 428 { 429 sigset_t sigset; 430 431 sigemptyset(&sigset); 432 sigaddset(&sigset, SIGALRM); 433 434 pthread_sigmask(SIG_BLOCK, &sigset, NULL); 435 436 signal(SIGKVMEXIT, kvm_cpu_signal_handler); 437 signal(SIGKVMPAUSE, kvm_cpu_signal_handler); 438 439 kvm_cpu__setup_cpuid(cpu); 440 kvm_cpu__reset_vcpu(cpu); 441 442 for (;;) { 443 if (cpu->paused) { 444 kvm__notify_paused(); 445 cpu->paused = 0; 446 } 447 448 kvm_cpu__run(cpu); 449 450 switch (cpu->kvm_run->exit_reason) { 451 case KVM_EXIT_UNKNOWN: 452 break; 453 case KVM_EXIT_DEBUG: 454 kvm_cpu__show_registers(cpu); 455 kvm_cpu__show_code(cpu); 456 break; 457 case KVM_EXIT_IO: { 458 bool ret; 459 460 ret = kvm__emulate_io(cpu->kvm, 461 cpu->kvm_run->io.port, 462 (u8 *)cpu->kvm_run + 463 cpu->kvm_run->io.data_offset, 464 cpu->kvm_run->io.direction, 465 cpu->kvm_run->io.size, 466 cpu->kvm_run->io.count); 467 468 if (!ret) 469 goto panic_kvm; 470 break; 471 } 472 case KVM_EXIT_MMIO: { 473 bool ret; 474 475 ret = kvm__emulate_mmio(cpu->kvm, 476 cpu->kvm_run->mmio.phys_addr, 477 cpu->kvm_run->mmio.data, 478 cpu->kvm_run->mmio.len, 479 cpu->kvm_run->mmio.is_write); 480 481 if (!ret) 482 goto panic_kvm; 483 break; 484 } 485 case KVM_EXIT_INTR: 486 if (cpu->is_running) 487 break; 488 goto exit_kvm; 489 case KVM_EXIT_SHUTDOWN: 490 goto exit_kvm; 491 default: 492 goto panic_kvm; 493 } 494 kvm_cpu__handle_coalesced_mmio(cpu); 495 } 496 497 exit_kvm: 498 return 0; 499 500 panic_kvm: 501 return 1; 502 } 503