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