1 /* 2 * x86/vmx.c : Framework for testing nested virtualization 3 * This is a framework to test nested VMX for KVM, which 4 * started as a project of GSoC 2013. All test cases should 5 * be located in x86/vmx_tests.c and framework related 6 * functions should be in this file. 7 * 8 * How to write test cases? 9 * Add callbacks of test suite in variant "vmx_tests". You can 10 * write: 11 * 1. init function used for initializing test suite 12 * 2. main function for codes running in L2 guest, 13 * 3. exit_handler to handle vmexit of L2 to L1 14 * 4. syscall handler to handle L2 syscall vmexit 15 * 5. vmenter fail handler to handle direct failure of vmenter 16 * 6. guest_regs is loaded when vmenter and saved when 17 * vmexit, you can read and set it in exit_handler 18 * If no special function is needed for a test suite, use 19 * coressponding basic_* functions as callback. More handlers 20 * can be added to "vmx_tests", see details of "struct vmx_test" 21 * and function test_run(). 22 * 23 * Currently, vmx test framework only set up one VCPU and one 24 * concurrent guest test environment with same paging for L2 and 25 * L1. For usage of EPT, only 1:1 mapped paging is used from VFN 26 * to PFN. 27 * 28 * Author : Arthur Chunqi Li <yzt356@gmail.com> 29 */ 30 31 #include "libcflat.h" 32 #include "processor.h" 33 #include "alloc_page.h" 34 #include "vm.h" 35 #include "desc.h" 36 #include "vmx.h" 37 #include "msr.h" 38 #include "smp.h" 39 40 u64 *vmxon_region; 41 struct vmcs *vmcs_root; 42 u32 vpid_cnt; 43 void *guest_stack, *guest_syscall_stack; 44 u32 ctrl_pin, ctrl_enter, ctrl_exit, ctrl_cpu[2]; 45 struct regs regs; 46 47 struct vmx_test *current; 48 49 #define MAX_TEST_TEARDOWN_STEPS 10 50 51 struct test_teardown_step { 52 test_teardown_func func; 53 void *data; 54 }; 55 56 static int teardown_count; 57 static struct test_teardown_step teardown_steps[MAX_TEST_TEARDOWN_STEPS]; 58 59 static test_guest_func v2_guest_main; 60 61 u64 hypercall_field; 62 bool launched; 63 static int matched; 64 static int guest_finished; 65 static int in_guest; 66 67 union vmx_basic basic; 68 union vmx_ctrl_msr ctrl_pin_rev; 69 union vmx_ctrl_msr ctrl_cpu_rev[2]; 70 union vmx_ctrl_msr ctrl_exit_rev; 71 union vmx_ctrl_msr ctrl_enter_rev; 72 union vmx_ept_vpid ept_vpid; 73 74 extern struct descriptor_table_ptr gdt64_desc; 75 extern struct descriptor_table_ptr idt_descr; 76 extern struct descriptor_table_ptr tss_descr; 77 extern void *vmx_return; 78 extern void *entry_sysenter; 79 extern void *guest_entry; 80 81 static volatile u32 stage; 82 83 static jmp_buf abort_target; 84 85 struct vmcs_field { 86 u64 mask; 87 u64 encoding; 88 }; 89 90 #define MASK(_bits) GENMASK_ULL((_bits) - 1, 0) 91 #define MASK_NATURAL MASK(sizeof(unsigned long) * 8) 92 93 static struct vmcs_field vmcs_fields[] = { 94 { MASK(16), VPID }, 95 { MASK(16), PINV }, 96 { MASK(16), EPTP_IDX }, 97 98 { MASK(16), GUEST_SEL_ES }, 99 { MASK(16), GUEST_SEL_CS }, 100 { MASK(16), GUEST_SEL_SS }, 101 { MASK(16), GUEST_SEL_DS }, 102 { MASK(16), GUEST_SEL_FS }, 103 { MASK(16), GUEST_SEL_GS }, 104 { MASK(16), GUEST_SEL_LDTR }, 105 { MASK(16), GUEST_SEL_TR }, 106 { MASK(16), GUEST_INT_STATUS }, 107 108 { MASK(16), HOST_SEL_ES }, 109 { MASK(16), HOST_SEL_CS }, 110 { MASK(16), HOST_SEL_SS }, 111 { MASK(16), HOST_SEL_DS }, 112 { MASK(16), HOST_SEL_FS }, 113 { MASK(16), HOST_SEL_GS }, 114 { MASK(16), HOST_SEL_TR }, 115 116 { MASK(64), IO_BITMAP_A }, 117 { MASK(64), IO_BITMAP_B }, 118 { MASK(64), MSR_BITMAP }, 119 { MASK(64), EXIT_MSR_ST_ADDR }, 120 { MASK(64), EXIT_MSR_LD_ADDR }, 121 { MASK(64), ENTER_MSR_LD_ADDR }, 122 { MASK(64), VMCS_EXEC_PTR }, 123 { MASK(64), TSC_OFFSET }, 124 { MASK(64), APIC_VIRT_ADDR }, 125 { MASK(64), APIC_ACCS_ADDR }, 126 { MASK(64), EPTP }, 127 128 { 0 /* read-only */, INFO_PHYS_ADDR }, 129 130 { MASK(64), VMCS_LINK_PTR }, 131 { MASK(64), GUEST_DEBUGCTL }, 132 { MASK(64), GUEST_EFER }, 133 { MASK(64), GUEST_PAT }, 134 { MASK(64), GUEST_PERF_GLOBAL_CTRL }, 135 { MASK(64), GUEST_PDPTE }, 136 137 { MASK(64), HOST_PAT }, 138 { MASK(64), HOST_EFER }, 139 { MASK(64), HOST_PERF_GLOBAL_CTRL }, 140 141 { MASK(32), PIN_CONTROLS }, 142 { MASK(32), CPU_EXEC_CTRL0 }, 143 { MASK(32), EXC_BITMAP }, 144 { MASK(32), PF_ERROR_MASK }, 145 { MASK(32), PF_ERROR_MATCH }, 146 { MASK(32), CR3_TARGET_COUNT }, 147 { MASK(32), EXI_CONTROLS }, 148 { MASK(32), EXI_MSR_ST_CNT }, 149 { MASK(32), EXI_MSR_LD_CNT }, 150 { MASK(32), ENT_CONTROLS }, 151 { MASK(32), ENT_MSR_LD_CNT }, 152 { MASK(32), ENT_INTR_INFO }, 153 { MASK(32), ENT_INTR_ERROR }, 154 { MASK(32), ENT_INST_LEN }, 155 { MASK(32), TPR_THRESHOLD }, 156 { MASK(32), CPU_EXEC_CTRL1 }, 157 158 { 0 /* read-only */, VMX_INST_ERROR }, 159 { 0 /* read-only */, EXI_REASON }, 160 { 0 /* read-only */, EXI_INTR_INFO }, 161 { 0 /* read-only */, EXI_INTR_ERROR }, 162 { 0 /* read-only */, IDT_VECT_INFO }, 163 { 0 /* read-only */, IDT_VECT_ERROR }, 164 { 0 /* read-only */, EXI_INST_LEN }, 165 { 0 /* read-only */, EXI_INST_INFO }, 166 167 { MASK(32), GUEST_LIMIT_ES }, 168 { MASK(32), GUEST_LIMIT_CS }, 169 { MASK(32), GUEST_LIMIT_SS }, 170 { MASK(32), GUEST_LIMIT_DS }, 171 { MASK(32), GUEST_LIMIT_FS }, 172 { MASK(32), GUEST_LIMIT_GS }, 173 { MASK(32), GUEST_LIMIT_LDTR }, 174 { MASK(32), GUEST_LIMIT_TR }, 175 { MASK(32), GUEST_LIMIT_GDTR }, 176 { MASK(32), GUEST_LIMIT_IDTR }, 177 { 0x1d0ff, GUEST_AR_ES }, 178 { 0x1f0ff, GUEST_AR_CS }, 179 { 0x1d0ff, GUEST_AR_SS }, 180 { 0x1d0ff, GUEST_AR_DS }, 181 { 0x1d0ff, GUEST_AR_FS }, 182 { 0x1d0ff, GUEST_AR_GS }, 183 { 0x1d0ff, GUEST_AR_LDTR }, 184 { 0x1d0ff, GUEST_AR_TR }, 185 { MASK(32), GUEST_INTR_STATE }, 186 { MASK(32), GUEST_ACTV_STATE }, 187 { MASK(32), GUEST_SMBASE }, 188 { MASK(32), GUEST_SYSENTER_CS }, 189 { MASK(32), PREEMPT_TIMER_VALUE }, 190 191 { MASK(32), HOST_SYSENTER_CS }, 192 193 { MASK_NATURAL, CR0_MASK }, 194 { MASK_NATURAL, CR4_MASK }, 195 { MASK_NATURAL, CR0_READ_SHADOW }, 196 { MASK_NATURAL, CR4_READ_SHADOW }, 197 { MASK_NATURAL, CR3_TARGET_0 }, 198 { MASK_NATURAL, CR3_TARGET_1 }, 199 { MASK_NATURAL, CR3_TARGET_2 }, 200 { MASK_NATURAL, CR3_TARGET_3 }, 201 202 { 0 /* read-only */, EXI_QUALIFICATION }, 203 { 0 /* read-only */, IO_RCX }, 204 { 0 /* read-only */, IO_RSI }, 205 { 0 /* read-only */, IO_RDI }, 206 { 0 /* read-only */, IO_RIP }, 207 { 0 /* read-only */, GUEST_LINEAR_ADDRESS }, 208 209 { MASK_NATURAL, GUEST_CR0 }, 210 { MASK_NATURAL, GUEST_CR3 }, 211 { MASK_NATURAL, GUEST_CR4 }, 212 { MASK_NATURAL, GUEST_BASE_ES }, 213 { MASK_NATURAL, GUEST_BASE_CS }, 214 { MASK_NATURAL, GUEST_BASE_SS }, 215 { MASK_NATURAL, GUEST_BASE_DS }, 216 { MASK_NATURAL, GUEST_BASE_FS }, 217 { MASK_NATURAL, GUEST_BASE_GS }, 218 { MASK_NATURAL, GUEST_BASE_LDTR }, 219 { MASK_NATURAL, GUEST_BASE_TR }, 220 { MASK_NATURAL, GUEST_BASE_GDTR }, 221 { MASK_NATURAL, GUEST_BASE_IDTR }, 222 { MASK_NATURAL, GUEST_DR7 }, 223 { MASK_NATURAL, GUEST_RSP }, 224 { MASK_NATURAL, GUEST_RIP }, 225 { MASK_NATURAL, GUEST_RFLAGS }, 226 { MASK_NATURAL, GUEST_PENDING_DEBUG }, 227 { MASK_NATURAL, GUEST_SYSENTER_ESP }, 228 { MASK_NATURAL, GUEST_SYSENTER_EIP }, 229 230 { MASK_NATURAL, HOST_CR0 }, 231 { MASK_NATURAL, HOST_CR3 }, 232 { MASK_NATURAL, HOST_CR4 }, 233 { MASK_NATURAL, HOST_BASE_FS }, 234 { MASK_NATURAL, HOST_BASE_GS }, 235 { MASK_NATURAL, HOST_BASE_TR }, 236 { MASK_NATURAL, HOST_BASE_GDTR }, 237 { MASK_NATURAL, HOST_BASE_IDTR }, 238 { MASK_NATURAL, HOST_SYSENTER_ESP }, 239 { MASK_NATURAL, HOST_SYSENTER_EIP }, 240 { MASK_NATURAL, HOST_RSP }, 241 { MASK_NATURAL, HOST_RIP }, 242 }; 243 244 static inline u64 vmcs_field_value(struct vmcs_field *f, u8 cookie) 245 { 246 u64 value; 247 248 /* Incorporate the cookie and the field encoding into the value. */ 249 value = cookie; 250 value |= (f->encoding << 8); 251 value |= 0xdeadbeefull << 32; 252 253 return value & f->mask; 254 } 255 256 static void set_vmcs_field(struct vmcs_field *f, u8 cookie) 257 { 258 vmcs_write(f->encoding, vmcs_field_value(f, cookie)); 259 } 260 261 static bool check_vmcs_field(struct vmcs_field *f, u8 cookie) 262 { 263 u64 expected; 264 u64 actual; 265 int ret; 266 267 ret = vmcs_read_checking(f->encoding, &actual); 268 assert(!(ret & X86_EFLAGS_CF)); 269 /* Skip VMCS fields that aren't recognized by the CPU */ 270 if (ret & X86_EFLAGS_ZF) 271 return true; 272 273 expected = vmcs_field_value(f, cookie); 274 actual &= f->mask; 275 276 if (expected == actual) 277 return true; 278 279 printf("FAIL: VMWRITE/VMREAD %lx (expected: %lx, actual: %lx)\n", 280 f->encoding, (unsigned long) expected, (unsigned long) actual); 281 282 return false; 283 } 284 285 static void set_all_vmcs_fields(u8 cookie) 286 { 287 int i; 288 289 for (i = 0; i < ARRAY_SIZE(vmcs_fields); i++) 290 set_vmcs_field(&vmcs_fields[i], cookie); 291 } 292 293 static bool check_all_vmcs_fields(u8 cookie) 294 { 295 bool pass = true; 296 int i; 297 298 for (i = 0; i < ARRAY_SIZE(vmcs_fields); i++) { 299 if (!check_vmcs_field(&vmcs_fields[i], cookie)) 300 pass = false; 301 } 302 303 return pass; 304 } 305 306 void test_vmwrite_vmread(void) 307 { 308 struct vmcs *vmcs = alloc_page(); 309 310 memset(vmcs, 0, PAGE_SIZE); 311 vmcs->revision_id = basic.revision; 312 assert(!vmcs_clear(vmcs)); 313 assert(!make_vmcs_current(vmcs)); 314 315 set_all_vmcs_fields(0x42); 316 report("VMWRITE/VMREAD", check_all_vmcs_fields(0x42)); 317 318 assert(!vmcs_clear(vmcs)); 319 free_page(vmcs); 320 } 321 322 void test_vmcs_high(void) 323 { 324 struct vmcs *vmcs = alloc_page(); 325 326 memset(vmcs, 0, PAGE_SIZE); 327 vmcs->revision_id = basic.revision; 328 assert(!vmcs_clear(vmcs)); 329 assert(!make_vmcs_current(vmcs)); 330 331 vmcs_write(TSC_OFFSET, 0x0123456789ABCDEFull); 332 report("VMREAD TSC_OFFSET after VMWRITE TSC_OFFSET", 333 vmcs_read(TSC_OFFSET) == 0x0123456789ABCDEFull); 334 report("VMREAD TSC_OFFSET_HI after VMWRITE TSC_OFFSET", 335 vmcs_read(TSC_OFFSET_HI) == 0x01234567ull); 336 vmcs_write(TSC_OFFSET_HI, 0x76543210ul); 337 report("VMREAD TSC_OFFSET_HI after VMWRITE TSC_OFFSET_HI", 338 vmcs_read(TSC_OFFSET_HI) == 0x76543210ul); 339 report("VMREAD TSC_OFFSET after VMWRITE TSC_OFFSET_HI", 340 vmcs_read(TSC_OFFSET) == 0x7654321089ABCDEFull); 341 342 assert(!vmcs_clear(vmcs)); 343 free_page(vmcs); 344 } 345 346 void test_vmcs_lifecycle(void) 347 { 348 struct vmcs *vmcs[2] = {}; 349 int i; 350 351 for (i = 0; i < ARRAY_SIZE(vmcs); i++) { 352 vmcs[i] = alloc_page(); 353 memset(vmcs[i], 0, PAGE_SIZE); 354 vmcs[i]->revision_id = basic.revision; 355 } 356 357 #define VMPTRLD(_i) do { \ 358 assert(_i < ARRAY_SIZE(vmcs)); \ 359 assert(!make_vmcs_current(vmcs[_i])); \ 360 printf("VMPTRLD VMCS%d\n", (_i)); \ 361 } while (0) 362 363 #define VMCLEAR(_i) do { \ 364 assert(_i < ARRAY_SIZE(vmcs)); \ 365 assert(!vmcs_clear(vmcs[_i])); \ 366 printf("VMCLEAR VMCS%d\n", (_i)); \ 367 } while (0) 368 369 VMCLEAR(0); 370 VMPTRLD(0); 371 set_all_vmcs_fields(0); 372 report("current:VMCS0 active:[VMCS0]", check_all_vmcs_fields(0)); 373 374 VMCLEAR(0); 375 VMPTRLD(0); 376 report("current:VMCS0 active:[VMCS0]", check_all_vmcs_fields(0)); 377 378 VMCLEAR(1); 379 report("current:VMCS0 active:[VMCS0]", check_all_vmcs_fields(0)); 380 381 VMPTRLD(1); 382 set_all_vmcs_fields(1); 383 report("current:VMCS1 active:[VMCS0,VCMS1]", check_all_vmcs_fields(1)); 384 385 VMPTRLD(0); 386 report("current:VMCS0 active:[VMCS0,VCMS1]", check_all_vmcs_fields(0)); 387 VMPTRLD(1); 388 report("current:VMCS1 active:[VMCS0,VCMS1]", check_all_vmcs_fields(1)); 389 VMPTRLD(1); 390 report("current:VMCS1 active:[VMCS0,VCMS1]", check_all_vmcs_fields(1)); 391 392 VMCLEAR(0); 393 report("current:VMCS1 active:[VCMS1]", check_all_vmcs_fields(1)); 394 395 /* VMPTRLD should not erase VMWRITEs to the current VMCS */ 396 set_all_vmcs_fields(2); 397 VMPTRLD(1); 398 report("current:VMCS1 active:[VCMS1]", check_all_vmcs_fields(2)); 399 400 for (i = 0; i < ARRAY_SIZE(vmcs); i++) { 401 VMCLEAR(i); 402 free_page(vmcs[i]); 403 } 404 405 #undef VMPTRLD 406 #undef VMCLEAR 407 } 408 409 void vmx_set_test_stage(u32 s) 410 { 411 barrier(); 412 stage = s; 413 barrier(); 414 } 415 416 u32 vmx_get_test_stage(void) 417 { 418 u32 s; 419 420 barrier(); 421 s = stage; 422 barrier(); 423 return s; 424 } 425 426 void vmx_inc_test_stage(void) 427 { 428 barrier(); 429 stage++; 430 barrier(); 431 } 432 433 /* entry_sysenter */ 434 asm( 435 ".align 4, 0x90\n\t" 436 ".globl entry_sysenter\n\t" 437 "entry_sysenter:\n\t" 438 SAVE_GPR 439 " and $0xf, %rax\n\t" 440 " mov %rax, %rdi\n\t" 441 " call syscall_handler\n\t" 442 LOAD_GPR 443 " vmresume\n\t" 444 ); 445 446 static void __attribute__((__used__)) syscall_handler(u64 syscall_no) 447 { 448 if (current->syscall_handler) 449 current->syscall_handler(syscall_no); 450 } 451 452 static const char * const exit_reason_descriptions[] = { 453 [VMX_EXC_NMI] = "VMX_EXC_NMI", 454 [VMX_EXTINT] = "VMX_EXTINT", 455 [VMX_TRIPLE_FAULT] = "VMX_TRIPLE_FAULT", 456 [VMX_INIT] = "VMX_INIT", 457 [VMX_SIPI] = "VMX_SIPI", 458 [VMX_SMI_IO] = "VMX_SMI_IO", 459 [VMX_SMI_OTHER] = "VMX_SMI_OTHER", 460 [VMX_INTR_WINDOW] = "VMX_INTR_WINDOW", 461 [VMX_NMI_WINDOW] = "VMX_NMI_WINDOW", 462 [VMX_TASK_SWITCH] = "VMX_TASK_SWITCH", 463 [VMX_CPUID] = "VMX_CPUID", 464 [VMX_GETSEC] = "VMX_GETSEC", 465 [VMX_HLT] = "VMX_HLT", 466 [VMX_INVD] = "VMX_INVD", 467 [VMX_INVLPG] = "VMX_INVLPG", 468 [VMX_RDPMC] = "VMX_RDPMC", 469 [VMX_RDTSC] = "VMX_RDTSC", 470 [VMX_RSM] = "VMX_RSM", 471 [VMX_VMCALL] = "VMX_VMCALL", 472 [VMX_VMCLEAR] = "VMX_VMCLEAR", 473 [VMX_VMLAUNCH] = "VMX_VMLAUNCH", 474 [VMX_VMPTRLD] = "VMX_VMPTRLD", 475 [VMX_VMPTRST] = "VMX_VMPTRST", 476 [VMX_VMREAD] = "VMX_VMREAD", 477 [VMX_VMRESUME] = "VMX_VMRESUME", 478 [VMX_VMWRITE] = "VMX_VMWRITE", 479 [VMX_VMXOFF] = "VMX_VMXOFF", 480 [VMX_VMXON] = "VMX_VMXON", 481 [VMX_CR] = "VMX_CR", 482 [VMX_DR] = "VMX_DR", 483 [VMX_IO] = "VMX_IO", 484 [VMX_RDMSR] = "VMX_RDMSR", 485 [VMX_WRMSR] = "VMX_WRMSR", 486 [VMX_FAIL_STATE] = "VMX_FAIL_STATE", 487 [VMX_FAIL_MSR] = "VMX_FAIL_MSR", 488 [VMX_MWAIT] = "VMX_MWAIT", 489 [VMX_MTF] = "VMX_MTF", 490 [VMX_MONITOR] = "VMX_MONITOR", 491 [VMX_PAUSE] = "VMX_PAUSE", 492 [VMX_FAIL_MCHECK] = "VMX_FAIL_MCHECK", 493 [VMX_TPR_THRESHOLD] = "VMX_TPR_THRESHOLD", 494 [VMX_APIC_ACCESS] = "VMX_APIC_ACCESS", 495 [VMX_GDTR_IDTR] = "VMX_GDTR_IDTR", 496 [VMX_LDTR_TR] = "VMX_LDTR_TR", 497 [VMX_EPT_VIOLATION] = "VMX_EPT_VIOLATION", 498 [VMX_EPT_MISCONFIG] = "VMX_EPT_MISCONFIG", 499 [VMX_INVEPT] = "VMX_INVEPT", 500 [VMX_PREEMPT] = "VMX_PREEMPT", 501 [VMX_INVVPID] = "VMX_INVVPID", 502 [VMX_WBINVD] = "VMX_WBINVD", 503 [VMX_XSETBV] = "VMX_XSETBV", 504 [VMX_APIC_WRITE] = "VMX_APIC_WRITE", 505 [VMX_RDRAND] = "VMX_RDRAND", 506 [VMX_INVPCID] = "VMX_INVPCID", 507 [VMX_VMFUNC] = "VMX_VMFUNC", 508 [VMX_RDSEED] = "VMX_RDSEED", 509 [VMX_PML_FULL] = "VMX_PML_FULL", 510 [VMX_XSAVES] = "VMX_XSAVES", 511 [VMX_XRSTORS] = "VMX_XRSTORS", 512 }; 513 514 const char *exit_reason_description(u64 reason) 515 { 516 if (reason >= ARRAY_SIZE(exit_reason_descriptions)) 517 return "(unknown)"; 518 return exit_reason_descriptions[reason] ? : "(unused)"; 519 } 520 521 void print_vmexit_info() 522 { 523 u64 guest_rip, guest_rsp; 524 ulong reason = vmcs_read(EXI_REASON) & 0xff; 525 ulong exit_qual = vmcs_read(EXI_QUALIFICATION); 526 guest_rip = vmcs_read(GUEST_RIP); 527 guest_rsp = vmcs_read(GUEST_RSP); 528 printf("VMEXIT info:\n"); 529 printf("\tvmexit reason = %ld\n", reason); 530 printf("\texit qualification = %#lx\n", exit_qual); 531 printf("\tBit 31 of reason = %lx\n", (vmcs_read(EXI_REASON) >> 31) & 1); 532 printf("\tguest_rip = %#lx\n", guest_rip); 533 printf("\tRAX=%#lx RBX=%#lx RCX=%#lx RDX=%#lx\n", 534 regs.rax, regs.rbx, regs.rcx, regs.rdx); 535 printf("\tRSP=%#lx RBP=%#lx RSI=%#lx RDI=%#lx\n", 536 guest_rsp, regs.rbp, regs.rsi, regs.rdi); 537 printf("\tR8 =%#lx R9 =%#lx R10=%#lx R11=%#lx\n", 538 regs.r8, regs.r9, regs.r10, regs.r11); 539 printf("\tR12=%#lx R13=%#lx R14=%#lx R15=%#lx\n", 540 regs.r12, regs.r13, regs.r14, regs.r15); 541 } 542 543 void 544 print_vmentry_failure_info(struct vmentry_failure *failure) { 545 if (failure->early) { 546 printf("Early %s failure: ", failure->instr); 547 switch (failure->flags & VMX_ENTRY_FLAGS) { 548 case X86_EFLAGS_CF: 549 printf("current-VMCS pointer is not valid.\n"); 550 break; 551 case X86_EFLAGS_ZF: 552 printf("error number is %ld. See Intel 30.4.\n", 553 vmcs_read(VMX_INST_ERROR)); 554 break; 555 default: 556 printf("unexpected flags %lx!\n", failure->flags); 557 } 558 } else { 559 u64 reason = vmcs_read(EXI_REASON); 560 u64 qual = vmcs_read(EXI_QUALIFICATION); 561 562 printf("Non-early %s failure (reason=%#lx, qual=%#lx): ", 563 failure->instr, reason, qual); 564 565 switch (reason & 0xff) { 566 case VMX_FAIL_STATE: 567 printf("invalid guest state\n"); 568 break; 569 case VMX_FAIL_MSR: 570 printf("MSR loading\n"); 571 break; 572 case VMX_FAIL_MCHECK: 573 printf("machine-check event\n"); 574 break; 575 default: 576 printf("unexpected basic exit reason %ld\n", 577 reason & 0xff); 578 } 579 580 if (!(reason & VMX_ENTRY_FAILURE)) 581 printf("\tVMX_ENTRY_FAILURE BIT NOT SET!\n"); 582 583 if (reason & 0x7fff0000) 584 printf("\tRESERVED BITS SET!\n"); 585 } 586 } 587 588 /* 589 * VMCLEAR should ensures all VMCS state is flushed to the VMCS 590 * region in memory. 591 */ 592 static void test_vmclear_flushing(void) 593 { 594 struct vmcs *vmcs[3] = {}; 595 int i; 596 597 for (i = 0; i < ARRAY_SIZE(vmcs); i++) { 598 vmcs[i] = alloc_page(); 599 memset(vmcs[i], 0, PAGE_SIZE); 600 } 601 602 vmcs[0]->revision_id = basic.revision; 603 assert(!vmcs_clear(vmcs[0])); 604 assert(!make_vmcs_current(vmcs[0])); 605 set_all_vmcs_fields(0x86); 606 607 assert(!vmcs_clear(vmcs[0])); 608 memcpy(vmcs[1], vmcs[0], basic.size); 609 assert(!make_vmcs_current(vmcs[1])); 610 report("test vmclear flush (current VMCS)", check_all_vmcs_fields(0x86)); 611 612 set_all_vmcs_fields(0x87); 613 assert(!make_vmcs_current(vmcs[0])); 614 assert(!vmcs_clear(vmcs[1])); 615 memcpy(vmcs[2], vmcs[1], basic.size); 616 assert(!make_vmcs_current(vmcs[2])); 617 report("test vmclear flush (!current VMCS)", check_all_vmcs_fields(0x87)); 618 619 for (i = 0; i < ARRAY_SIZE(vmcs); i++) { 620 assert(!vmcs_clear(vmcs[i])); 621 free_page(vmcs[i]); 622 } 623 } 624 625 static void test_vmclear(void) 626 { 627 struct vmcs *tmp_root; 628 int width = cpuid_maxphyaddr(); 629 630 /* 631 * Note- The tests below do not necessarily have a 632 * valid VMCS, but that's ok since the invalid vmcs 633 * is only used for a specific test and is discarded 634 * without touching its contents 635 */ 636 637 /* Unaligned page access */ 638 tmp_root = (struct vmcs *)((intptr_t)vmcs_root + 1); 639 report("test vmclear with unaligned vmcs", 640 vmcs_clear(tmp_root) == 1); 641 642 /* gpa bits beyond physical address width are set*/ 643 tmp_root = (struct vmcs *)((intptr_t)vmcs_root | 644 ((u64)1 << (width+1))); 645 report("test vmclear with vmcs address bits set beyond physical address width", 646 vmcs_clear(tmp_root) == 1); 647 648 /* Pass VMXON region */ 649 tmp_root = (struct vmcs *)vmxon_region; 650 report("test vmclear with vmxon region", 651 vmcs_clear(tmp_root) == 1); 652 653 /* Valid VMCS */ 654 report("test vmclear with valid vmcs region", vmcs_clear(vmcs_root) == 0); 655 656 test_vmclear_flushing(); 657 } 658 659 static void __attribute__((__used__)) guest_main(void) 660 { 661 if (current->v2) 662 v2_guest_main(); 663 else 664 current->guest_main(); 665 } 666 667 /* guest_entry */ 668 asm( 669 ".align 4, 0x90\n\t" 670 ".globl entry_guest\n\t" 671 "guest_entry:\n\t" 672 " call guest_main\n\t" 673 " mov $1, %edi\n\t" 674 " call hypercall\n\t" 675 ); 676 677 /* EPT paging structure related functions */ 678 /* split_large_ept_entry: Split a 2M/1G large page into 512 smaller PTEs. 679 @ptep : large page table entry to split 680 @level : level of ptep (2 or 3) 681 */ 682 static void split_large_ept_entry(unsigned long *ptep, int level) 683 { 684 unsigned long *new_pt; 685 unsigned long gpa; 686 unsigned long pte; 687 unsigned long prototype; 688 int i; 689 690 pte = *ptep; 691 assert(pte & EPT_PRESENT); 692 assert(pte & EPT_LARGE_PAGE); 693 assert(level == 2 || level == 3); 694 695 new_pt = alloc_page(); 696 assert(new_pt); 697 memset(new_pt, 0, PAGE_SIZE); 698 699 prototype = pte & ~EPT_ADDR_MASK; 700 if (level == 2) 701 prototype &= ~EPT_LARGE_PAGE; 702 703 gpa = pte & EPT_ADDR_MASK; 704 for (i = 0; i < EPT_PGDIR_ENTRIES; i++) { 705 new_pt[i] = prototype | gpa; 706 gpa += 1ul << EPT_LEVEL_SHIFT(level - 1); 707 } 708 709 pte &= ~EPT_LARGE_PAGE; 710 pte &= ~EPT_ADDR_MASK; 711 pte |= virt_to_phys(new_pt); 712 713 *ptep = pte; 714 } 715 716 /* install_ept_entry : Install a page to a given level in EPT 717 @pml4 : addr of pml4 table 718 @pte_level : level of PTE to set 719 @guest_addr : physical address of guest 720 @pte : pte value to set 721 @pt_page : address of page table, NULL for a new page 722 */ 723 void install_ept_entry(unsigned long *pml4, 724 int pte_level, 725 unsigned long guest_addr, 726 unsigned long pte, 727 unsigned long *pt_page) 728 { 729 int level; 730 unsigned long *pt = pml4; 731 unsigned offset; 732 733 /* EPT only uses 48 bits of GPA. */ 734 assert(guest_addr < (1ul << 48)); 735 736 for (level = EPT_PAGE_LEVEL; level > pte_level; --level) { 737 offset = (guest_addr >> EPT_LEVEL_SHIFT(level)) 738 & EPT_PGDIR_MASK; 739 if (!(pt[offset] & (EPT_PRESENT))) { 740 unsigned long *new_pt = pt_page; 741 if (!new_pt) 742 new_pt = alloc_page(); 743 else 744 pt_page = 0; 745 memset(new_pt, 0, PAGE_SIZE); 746 pt[offset] = virt_to_phys(new_pt) 747 | EPT_RA | EPT_WA | EPT_EA; 748 } else if (pt[offset] & EPT_LARGE_PAGE) 749 split_large_ept_entry(&pt[offset], level); 750 pt = phys_to_virt(pt[offset] & EPT_ADDR_MASK); 751 } 752 offset = (guest_addr >> EPT_LEVEL_SHIFT(level)) & EPT_PGDIR_MASK; 753 pt[offset] = pte; 754 } 755 756 /* Map a page, @perm is the permission of the page */ 757 void install_ept(unsigned long *pml4, 758 unsigned long phys, 759 unsigned long guest_addr, 760 u64 perm) 761 { 762 install_ept_entry(pml4, 1, guest_addr, (phys & PAGE_MASK) | perm, 0); 763 } 764 765 /* Map a 1G-size page */ 766 void install_1g_ept(unsigned long *pml4, 767 unsigned long phys, 768 unsigned long guest_addr, 769 u64 perm) 770 { 771 install_ept_entry(pml4, 3, guest_addr, 772 (phys & PAGE_MASK) | perm | EPT_LARGE_PAGE, 0); 773 } 774 775 /* Map a 2M-size page */ 776 void install_2m_ept(unsigned long *pml4, 777 unsigned long phys, 778 unsigned long guest_addr, 779 u64 perm) 780 { 781 install_ept_entry(pml4, 2, guest_addr, 782 (phys & PAGE_MASK) | perm | EPT_LARGE_PAGE, 0); 783 } 784 785 /* setup_ept_range : Setup a range of 1:1 mapped page to EPT paging structure. 786 @start : start address of guest page 787 @len : length of address to be mapped 788 @map_1g : whether 1G page map is used 789 @map_2m : whether 2M page map is used 790 @perm : permission for every page 791 */ 792 void setup_ept_range(unsigned long *pml4, unsigned long start, 793 unsigned long len, int map_1g, int map_2m, u64 perm) 794 { 795 u64 phys = start; 796 u64 max = (u64)len + (u64)start; 797 798 if (map_1g) { 799 while (phys + PAGE_SIZE_1G <= max) { 800 install_1g_ept(pml4, phys, phys, perm); 801 phys += PAGE_SIZE_1G; 802 } 803 } 804 if (map_2m) { 805 while (phys + PAGE_SIZE_2M <= max) { 806 install_2m_ept(pml4, phys, phys, perm); 807 phys += PAGE_SIZE_2M; 808 } 809 } 810 while (phys + PAGE_SIZE <= max) { 811 install_ept(pml4, phys, phys, perm); 812 phys += PAGE_SIZE; 813 } 814 } 815 816 /* get_ept_pte : Get the PTE of a given level in EPT, 817 @level == 1 means get the latest level*/ 818 bool get_ept_pte(unsigned long *pml4, unsigned long guest_addr, int level, 819 unsigned long *pte) 820 { 821 int l; 822 unsigned long *pt = pml4, iter_pte; 823 unsigned offset; 824 825 assert(level >= 1 && level <= 4); 826 827 for (l = EPT_PAGE_LEVEL; ; --l) { 828 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 829 iter_pte = pt[offset]; 830 if (l == level) 831 break; 832 if (l < 4 && (iter_pte & EPT_LARGE_PAGE)) 833 return false; 834 if (!(iter_pte & (EPT_PRESENT))) 835 return false; 836 pt = (unsigned long *)(iter_pte & EPT_ADDR_MASK); 837 } 838 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 839 if (pte) 840 *pte = pt[offset]; 841 return true; 842 } 843 844 static void clear_ept_ad_pte(unsigned long *pml4, unsigned long guest_addr) 845 { 846 int l; 847 unsigned long *pt = pml4; 848 u64 pte; 849 unsigned offset; 850 851 for (l = EPT_PAGE_LEVEL; ; --l) { 852 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 853 pt[offset] &= ~(EPT_ACCESS_FLAG|EPT_DIRTY_FLAG); 854 pte = pt[offset]; 855 if (l == 1 || (l < 4 && (pte & EPT_LARGE_PAGE))) 856 break; 857 pt = (unsigned long *)(pte & EPT_ADDR_MASK); 858 } 859 } 860 861 /* clear_ept_ad : Clear EPT A/D bits for the page table walk and the 862 final GPA of a guest address. */ 863 void clear_ept_ad(unsigned long *pml4, u64 guest_cr3, 864 unsigned long guest_addr) 865 { 866 int l; 867 unsigned long *pt = (unsigned long *)guest_cr3, gpa; 868 u64 pte, offset_in_page; 869 unsigned offset; 870 871 for (l = EPT_PAGE_LEVEL; ; --l) { 872 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 873 874 clear_ept_ad_pte(pml4, (u64) &pt[offset]); 875 pte = pt[offset]; 876 if (l == 1 || (l < 4 && (pte & PT_PAGE_SIZE_MASK))) 877 break; 878 if (!(pte & PT_PRESENT_MASK)) 879 return; 880 pt = (unsigned long *)(pte & PT_ADDR_MASK); 881 } 882 883 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 884 offset_in_page = guest_addr & ((1 << EPT_LEVEL_SHIFT(l)) - 1); 885 gpa = (pt[offset] & PT_ADDR_MASK) | (guest_addr & offset_in_page); 886 clear_ept_ad_pte(pml4, gpa); 887 } 888 889 /* check_ept_ad : Check the content of EPT A/D bits for the page table 890 walk and the final GPA of a guest address. */ 891 void check_ept_ad(unsigned long *pml4, u64 guest_cr3, 892 unsigned long guest_addr, int expected_gpa_ad, 893 int expected_pt_ad) 894 { 895 int l; 896 unsigned long *pt = (unsigned long *)guest_cr3, gpa; 897 u64 ept_pte, pte, offset_in_page; 898 unsigned offset; 899 bool bad_pt_ad = false; 900 901 for (l = EPT_PAGE_LEVEL; ; --l) { 902 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 903 904 if (!get_ept_pte(pml4, (u64) &pt[offset], 1, &ept_pte)) { 905 printf("EPT - guest level %d page table is not mapped.\n", l); 906 return; 907 } 908 909 if (!bad_pt_ad) { 910 bad_pt_ad |= (ept_pte & (EPT_ACCESS_FLAG|EPT_DIRTY_FLAG)) != expected_pt_ad; 911 if (bad_pt_ad) 912 report("EPT - guest level %d page table A=%d/D=%d", 913 false, l, 914 !!(expected_pt_ad & EPT_ACCESS_FLAG), 915 !!(expected_pt_ad & EPT_DIRTY_FLAG)); 916 } 917 918 pte = pt[offset]; 919 if (l == 1 || (l < 4 && (pte & PT_PAGE_SIZE_MASK))) 920 break; 921 if (!(pte & PT_PRESENT_MASK)) 922 return; 923 pt = (unsigned long *)(pte & PT_ADDR_MASK); 924 } 925 926 if (!bad_pt_ad) 927 report("EPT - guest page table structures A=%d/D=%d", 928 true, 929 !!(expected_pt_ad & EPT_ACCESS_FLAG), 930 !!(expected_pt_ad & EPT_DIRTY_FLAG)); 931 932 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 933 offset_in_page = guest_addr & ((1 << EPT_LEVEL_SHIFT(l)) - 1); 934 gpa = (pt[offset] & PT_ADDR_MASK) | (guest_addr & offset_in_page); 935 936 if (!get_ept_pte(pml4, gpa, 1, &ept_pte)) { 937 report("EPT - guest physical address is not mapped", false); 938 return; 939 } 940 report("EPT - guest physical address A=%d/D=%d", 941 (ept_pte & (EPT_ACCESS_FLAG|EPT_DIRTY_FLAG)) == expected_gpa_ad, 942 !!(expected_gpa_ad & EPT_ACCESS_FLAG), 943 !!(expected_gpa_ad & EPT_DIRTY_FLAG)); 944 } 945 946 947 void ept_sync(int type, u64 eptp) 948 { 949 switch (type) { 950 case INVEPT_SINGLE: 951 if (ept_vpid.val & EPT_CAP_INVEPT_SINGLE) { 952 invept(INVEPT_SINGLE, eptp); 953 break; 954 } 955 /* else fall through */ 956 case INVEPT_GLOBAL: 957 if (ept_vpid.val & EPT_CAP_INVEPT_ALL) { 958 invept(INVEPT_GLOBAL, eptp); 959 break; 960 } 961 /* else fall through */ 962 default: 963 printf("WARNING: invept is not supported!\n"); 964 } 965 } 966 967 void set_ept_pte(unsigned long *pml4, unsigned long guest_addr, 968 int level, u64 pte_val) 969 { 970 int l; 971 unsigned long *pt = pml4; 972 unsigned offset; 973 974 assert(level >= 1 && level <= 4); 975 976 for (l = EPT_PAGE_LEVEL; ; --l) { 977 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 978 if (l == level) 979 break; 980 assert(pt[offset] & EPT_PRESENT); 981 pt = (unsigned long *)(pt[offset] & EPT_ADDR_MASK); 982 } 983 offset = (guest_addr >> EPT_LEVEL_SHIFT(l)) & EPT_PGDIR_MASK; 984 pt[offset] = pte_val; 985 } 986 987 bool ept_2m_supported(void) 988 { 989 return ept_vpid.val & EPT_CAP_2M_PAGE; 990 } 991 992 bool ept_1g_supported(void) 993 { 994 return ept_vpid.val & EPT_CAP_1G_PAGE; 995 } 996 997 bool ept_huge_pages_supported(int level) 998 { 999 if (level == 2) 1000 return ept_2m_supported(); 1001 else if (level == 3) 1002 return ept_1g_supported(); 1003 else 1004 return false; 1005 } 1006 1007 bool ept_execute_only_supported(void) 1008 { 1009 return ept_vpid.val & EPT_CAP_WT; 1010 } 1011 1012 bool ept_ad_bits_supported(void) 1013 { 1014 return ept_vpid.val & EPT_CAP_AD_FLAG; 1015 } 1016 1017 void vpid_sync(int type, u16 vpid) 1018 { 1019 switch(type) { 1020 case INVVPID_CONTEXT_GLOBAL: 1021 if (ept_vpid.val & VPID_CAP_INVVPID_CXTGLB) { 1022 invvpid(INVVPID_CONTEXT_GLOBAL, vpid, 0); 1023 break; 1024 } 1025 case INVVPID_ALL: 1026 if (ept_vpid.val & VPID_CAP_INVVPID_ALL) { 1027 invvpid(INVVPID_ALL, vpid, 0); 1028 break; 1029 } 1030 default: 1031 printf("WARNING: invvpid is not supported\n"); 1032 } 1033 } 1034 1035 static void init_vmcs_ctrl(void) 1036 { 1037 /* 26.2 CHECKS ON VMX CONTROLS AND HOST-STATE AREA */ 1038 /* 26.2.1.1 */ 1039 vmcs_write(PIN_CONTROLS, ctrl_pin); 1040 /* Disable VMEXIT of IO instruction */ 1041 vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu[0]); 1042 if (ctrl_cpu_rev[0].set & CPU_SECONDARY) { 1043 ctrl_cpu[1] = (ctrl_cpu[1] | ctrl_cpu_rev[1].set) & 1044 ctrl_cpu_rev[1].clr; 1045 vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu[1]); 1046 } 1047 vmcs_write(CR3_TARGET_COUNT, 0); 1048 vmcs_write(VPID, ++vpid_cnt); 1049 } 1050 1051 static void init_vmcs_host(void) 1052 { 1053 /* 26.2 CHECKS ON VMX CONTROLS AND HOST-STATE AREA */ 1054 /* 26.2.1.2 */ 1055 vmcs_write(HOST_EFER, rdmsr(MSR_EFER)); 1056 1057 /* 26.2.1.3 */ 1058 vmcs_write(ENT_CONTROLS, ctrl_enter); 1059 vmcs_write(EXI_CONTROLS, ctrl_exit); 1060 1061 /* 26.2.2 */ 1062 vmcs_write(HOST_CR0, read_cr0()); 1063 vmcs_write(HOST_CR3, read_cr3()); 1064 vmcs_write(HOST_CR4, read_cr4()); 1065 vmcs_write(HOST_SYSENTER_EIP, (u64)(&entry_sysenter)); 1066 vmcs_write(HOST_SYSENTER_CS, KERNEL_CS); 1067 1068 /* 26.2.3 */ 1069 vmcs_write(HOST_SEL_CS, KERNEL_CS); 1070 vmcs_write(HOST_SEL_SS, KERNEL_DS); 1071 vmcs_write(HOST_SEL_DS, KERNEL_DS); 1072 vmcs_write(HOST_SEL_ES, KERNEL_DS); 1073 vmcs_write(HOST_SEL_FS, KERNEL_DS); 1074 vmcs_write(HOST_SEL_GS, KERNEL_DS); 1075 vmcs_write(HOST_SEL_TR, TSS_MAIN); 1076 vmcs_write(HOST_BASE_TR, tss_descr.base); 1077 vmcs_write(HOST_BASE_GDTR, gdt64_desc.base); 1078 vmcs_write(HOST_BASE_IDTR, idt_descr.base); 1079 vmcs_write(HOST_BASE_FS, 0); 1080 vmcs_write(HOST_BASE_GS, 0); 1081 1082 /* Set other vmcs area */ 1083 vmcs_write(PF_ERROR_MASK, 0); 1084 vmcs_write(PF_ERROR_MATCH, 0); 1085 vmcs_write(VMCS_LINK_PTR, ~0ul); 1086 vmcs_write(VMCS_LINK_PTR_HI, ~0ul); 1087 vmcs_write(HOST_RIP, (u64)(&vmx_return)); 1088 } 1089 1090 static void init_vmcs_guest(void) 1091 { 1092 /* 26.3 CHECKING AND LOADING GUEST STATE */ 1093 ulong guest_cr0, guest_cr4, guest_cr3; 1094 /* 26.3.1.1 */ 1095 guest_cr0 = read_cr0(); 1096 guest_cr4 = read_cr4(); 1097 guest_cr3 = read_cr3(); 1098 if (ctrl_enter & ENT_GUEST_64) { 1099 guest_cr0 |= X86_CR0_PG; 1100 guest_cr4 |= X86_CR4_PAE; 1101 } 1102 if ((ctrl_enter & ENT_GUEST_64) == 0) 1103 guest_cr4 &= (~X86_CR4_PCIDE); 1104 if (guest_cr0 & X86_CR0_PG) 1105 guest_cr0 |= X86_CR0_PE; 1106 vmcs_write(GUEST_CR0, guest_cr0); 1107 vmcs_write(GUEST_CR3, guest_cr3); 1108 vmcs_write(GUEST_CR4, guest_cr4); 1109 vmcs_write(GUEST_SYSENTER_CS, KERNEL_CS); 1110 vmcs_write(GUEST_SYSENTER_ESP, 1111 (u64)(guest_syscall_stack + PAGE_SIZE - 1)); 1112 vmcs_write(GUEST_SYSENTER_EIP, (u64)(&entry_sysenter)); 1113 vmcs_write(GUEST_DR7, 0); 1114 vmcs_write(GUEST_EFER, rdmsr(MSR_EFER)); 1115 1116 /* 26.3.1.2 */ 1117 vmcs_write(GUEST_SEL_CS, KERNEL_CS); 1118 vmcs_write(GUEST_SEL_SS, KERNEL_DS); 1119 vmcs_write(GUEST_SEL_DS, KERNEL_DS); 1120 vmcs_write(GUEST_SEL_ES, KERNEL_DS); 1121 vmcs_write(GUEST_SEL_FS, KERNEL_DS); 1122 vmcs_write(GUEST_SEL_GS, KERNEL_DS); 1123 vmcs_write(GUEST_SEL_TR, TSS_MAIN); 1124 vmcs_write(GUEST_SEL_LDTR, 0); 1125 1126 vmcs_write(GUEST_BASE_CS, 0); 1127 vmcs_write(GUEST_BASE_ES, 0); 1128 vmcs_write(GUEST_BASE_SS, 0); 1129 vmcs_write(GUEST_BASE_DS, 0); 1130 vmcs_write(GUEST_BASE_FS, 0); 1131 vmcs_write(GUEST_BASE_GS, 0); 1132 vmcs_write(GUEST_BASE_TR, tss_descr.base); 1133 vmcs_write(GUEST_BASE_LDTR, 0); 1134 1135 vmcs_write(GUEST_LIMIT_CS, 0xFFFFFFFF); 1136 vmcs_write(GUEST_LIMIT_DS, 0xFFFFFFFF); 1137 vmcs_write(GUEST_LIMIT_ES, 0xFFFFFFFF); 1138 vmcs_write(GUEST_LIMIT_SS, 0xFFFFFFFF); 1139 vmcs_write(GUEST_LIMIT_FS, 0xFFFFFFFF); 1140 vmcs_write(GUEST_LIMIT_GS, 0xFFFFFFFF); 1141 vmcs_write(GUEST_LIMIT_LDTR, 0xffff); 1142 vmcs_write(GUEST_LIMIT_TR, tss_descr.limit); 1143 1144 vmcs_write(GUEST_AR_CS, 0xa09b); 1145 vmcs_write(GUEST_AR_DS, 0xc093); 1146 vmcs_write(GUEST_AR_ES, 0xc093); 1147 vmcs_write(GUEST_AR_FS, 0xc093); 1148 vmcs_write(GUEST_AR_GS, 0xc093); 1149 vmcs_write(GUEST_AR_SS, 0xc093); 1150 vmcs_write(GUEST_AR_LDTR, 0x82); 1151 vmcs_write(GUEST_AR_TR, 0x8b); 1152 1153 /* 26.3.1.3 */ 1154 vmcs_write(GUEST_BASE_GDTR, gdt64_desc.base); 1155 vmcs_write(GUEST_BASE_IDTR, idt_descr.base); 1156 vmcs_write(GUEST_LIMIT_GDTR, gdt64_desc.limit); 1157 vmcs_write(GUEST_LIMIT_IDTR, idt_descr.limit); 1158 1159 /* 26.3.1.4 */ 1160 vmcs_write(GUEST_RIP, (u64)(&guest_entry)); 1161 vmcs_write(GUEST_RSP, (u64)(guest_stack + PAGE_SIZE - 1)); 1162 vmcs_write(GUEST_RFLAGS, 0x2); 1163 1164 /* 26.3.1.5 */ 1165 vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE); 1166 vmcs_write(GUEST_INTR_STATE, 0); 1167 } 1168 1169 static int init_vmcs(struct vmcs **vmcs) 1170 { 1171 *vmcs = alloc_page(); 1172 memset(*vmcs, 0, PAGE_SIZE); 1173 (*vmcs)->revision_id = basic.revision; 1174 /* vmclear first to init vmcs */ 1175 if (vmcs_clear(*vmcs)) { 1176 printf("%s : vmcs_clear error\n", __func__); 1177 return 1; 1178 } 1179 1180 if (make_vmcs_current(*vmcs)) { 1181 printf("%s : make_vmcs_current error\n", __func__); 1182 return 1; 1183 } 1184 1185 /* All settings to pin/exit/enter/cpu 1186 control fields should be placed here */ 1187 ctrl_pin |= PIN_EXTINT | PIN_NMI | PIN_VIRT_NMI; 1188 ctrl_exit = EXI_LOAD_EFER | EXI_HOST_64; 1189 ctrl_enter = (ENT_LOAD_EFER | ENT_GUEST_64); 1190 /* DIsable IO instruction VMEXIT now */ 1191 ctrl_cpu[0] &= (~(CPU_IO | CPU_IO_BITMAP)); 1192 ctrl_cpu[1] = 0; 1193 1194 ctrl_pin = (ctrl_pin | ctrl_pin_rev.set) & ctrl_pin_rev.clr; 1195 ctrl_enter = (ctrl_enter | ctrl_enter_rev.set) & ctrl_enter_rev.clr; 1196 ctrl_exit = (ctrl_exit | ctrl_exit_rev.set) & ctrl_exit_rev.clr; 1197 ctrl_cpu[0] = (ctrl_cpu[0] | ctrl_cpu_rev[0].set) & ctrl_cpu_rev[0].clr; 1198 1199 init_vmcs_ctrl(); 1200 init_vmcs_host(); 1201 init_vmcs_guest(); 1202 return 0; 1203 } 1204 1205 static void init_vmx(void) 1206 { 1207 ulong fix_cr0_set, fix_cr0_clr; 1208 ulong fix_cr4_set, fix_cr4_clr; 1209 1210 vmxon_region = alloc_page(); 1211 memset(vmxon_region, 0, PAGE_SIZE); 1212 1213 fix_cr0_set = rdmsr(MSR_IA32_VMX_CR0_FIXED0); 1214 fix_cr0_clr = rdmsr(MSR_IA32_VMX_CR0_FIXED1); 1215 fix_cr4_set = rdmsr(MSR_IA32_VMX_CR4_FIXED0); 1216 fix_cr4_clr = rdmsr(MSR_IA32_VMX_CR4_FIXED1); 1217 basic.val = rdmsr(MSR_IA32_VMX_BASIC); 1218 ctrl_pin_rev.val = rdmsr(basic.ctrl ? MSR_IA32_VMX_TRUE_PIN 1219 : MSR_IA32_VMX_PINBASED_CTLS); 1220 ctrl_exit_rev.val = rdmsr(basic.ctrl ? MSR_IA32_VMX_TRUE_EXIT 1221 : MSR_IA32_VMX_EXIT_CTLS); 1222 ctrl_enter_rev.val = rdmsr(basic.ctrl ? MSR_IA32_VMX_TRUE_ENTRY 1223 : MSR_IA32_VMX_ENTRY_CTLS); 1224 ctrl_cpu_rev[0].val = rdmsr(basic.ctrl ? MSR_IA32_VMX_TRUE_PROC 1225 : MSR_IA32_VMX_PROCBASED_CTLS); 1226 if ((ctrl_cpu_rev[0].clr & CPU_SECONDARY) != 0) 1227 ctrl_cpu_rev[1].val = rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2); 1228 else 1229 ctrl_cpu_rev[1].val = 0; 1230 if ((ctrl_cpu_rev[1].clr & (CPU_EPT | CPU_VPID)) != 0) 1231 ept_vpid.val = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP); 1232 else 1233 ept_vpid.val = 0; 1234 1235 write_cr0((read_cr0() & fix_cr0_clr) | fix_cr0_set); 1236 write_cr4((read_cr4() & fix_cr4_clr) | fix_cr4_set | X86_CR4_VMXE); 1237 1238 *vmxon_region = basic.revision; 1239 1240 guest_stack = alloc_page(); 1241 memset(guest_stack, 0, PAGE_SIZE); 1242 guest_syscall_stack = alloc_page(); 1243 memset(guest_syscall_stack, 0, PAGE_SIZE); 1244 } 1245 1246 static void do_vmxon_off(void *data) 1247 { 1248 vmx_on(); 1249 vmx_off(); 1250 } 1251 1252 static void do_write_feature_control(void *data) 1253 { 1254 wrmsr(MSR_IA32_FEATURE_CONTROL, 0); 1255 } 1256 1257 static int test_vmx_feature_control(void) 1258 { 1259 u64 ia32_feature_control; 1260 bool vmx_enabled; 1261 1262 ia32_feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); 1263 vmx_enabled = ((ia32_feature_control & 0x5) == 0x5); 1264 if ((ia32_feature_control & 0x5) == 0x5) { 1265 printf("VMX enabled and locked by BIOS\n"); 1266 return 0; 1267 } else if (ia32_feature_control & 0x1) { 1268 printf("ERROR: VMX locked out by BIOS!?\n"); 1269 return 1; 1270 } 1271 1272 wrmsr(MSR_IA32_FEATURE_CONTROL, 0); 1273 report("test vmxon with FEATURE_CONTROL cleared", 1274 test_for_exception(GP_VECTOR, &do_vmxon_off, NULL)); 1275 1276 wrmsr(MSR_IA32_FEATURE_CONTROL, 0x4); 1277 report("test vmxon without FEATURE_CONTROL lock", 1278 test_for_exception(GP_VECTOR, &do_vmxon_off, NULL)); 1279 1280 wrmsr(MSR_IA32_FEATURE_CONTROL, 0x5); 1281 vmx_enabled = ((rdmsr(MSR_IA32_FEATURE_CONTROL) & 0x5) == 0x5); 1282 report("test enable VMX in FEATURE_CONTROL", vmx_enabled); 1283 1284 report("test FEATURE_CONTROL lock bit", 1285 test_for_exception(GP_VECTOR, &do_write_feature_control, NULL)); 1286 1287 return !vmx_enabled; 1288 } 1289 1290 static int test_vmxon(void) 1291 { 1292 int ret, ret1; 1293 u64 *tmp_region = vmxon_region; 1294 int width = cpuid_maxphyaddr(); 1295 1296 /* Unaligned page access */ 1297 vmxon_region = (u64 *)((intptr_t)vmxon_region + 1); 1298 ret1 = vmx_on(); 1299 report("test vmxon with unaligned vmxon region", ret1); 1300 if (!ret1) { 1301 ret = 1; 1302 goto out; 1303 } 1304 1305 /* gpa bits beyond physical address width are set*/ 1306 vmxon_region = (u64 *)((intptr_t)tmp_region | ((u64)1 << (width+1))); 1307 ret1 = vmx_on(); 1308 report("test vmxon with bits set beyond physical address width", ret1); 1309 if (!ret1) { 1310 ret = 1; 1311 goto out; 1312 } 1313 1314 /* invalid revision indentifier */ 1315 vmxon_region = tmp_region; 1316 *vmxon_region = 0xba9da9; 1317 ret1 = vmx_on(); 1318 report("test vmxon with invalid revision identifier", ret1); 1319 if (!ret1) { 1320 ret = 1; 1321 goto out; 1322 } 1323 1324 /* and finally a valid region */ 1325 *vmxon_region = basic.revision; 1326 ret = vmx_on(); 1327 report("test vmxon with valid vmxon region", !ret); 1328 1329 out: 1330 return ret; 1331 } 1332 1333 static void test_vmptrld(void) 1334 { 1335 struct vmcs *vmcs, *tmp_root; 1336 int width = cpuid_maxphyaddr(); 1337 1338 vmcs = alloc_page(); 1339 vmcs->revision_id = basic.revision; 1340 1341 /* Unaligned page access */ 1342 tmp_root = (struct vmcs *)((intptr_t)vmcs + 1); 1343 report("test vmptrld with unaligned vmcs", 1344 make_vmcs_current(tmp_root) == 1); 1345 1346 /* gpa bits beyond physical address width are set*/ 1347 tmp_root = (struct vmcs *)((intptr_t)vmcs | 1348 ((u64)1 << (width+1))); 1349 report("test vmptrld with vmcs address bits set beyond physical address width", 1350 make_vmcs_current(tmp_root) == 1); 1351 1352 /* Pass VMXON region */ 1353 make_vmcs_current(vmcs); 1354 tmp_root = (struct vmcs *)vmxon_region; 1355 report("test vmptrld with vmxon region", 1356 make_vmcs_current(tmp_root) == 1); 1357 report("test vmptrld with vmxon region vm-instruction error", 1358 vmcs_read(VMX_INST_ERROR) == VMXERR_VMPTRLD_VMXON_POINTER); 1359 1360 report("test vmptrld with valid vmcs region", make_vmcs_current(vmcs) == 0); 1361 } 1362 1363 static void test_vmptrst(void) 1364 { 1365 int ret; 1366 struct vmcs *vmcs1, *vmcs2; 1367 1368 vmcs1 = alloc_page(); 1369 memset(vmcs1, 0, PAGE_SIZE); 1370 init_vmcs(&vmcs1); 1371 ret = vmcs_save(&vmcs2); 1372 report("test vmptrst", (!ret) && (vmcs1 == vmcs2)); 1373 } 1374 1375 struct vmx_ctl_msr { 1376 const char *name; 1377 u32 index, true_index; 1378 u32 default1; 1379 } vmx_ctl_msr[] = { 1380 { "MSR_IA32_VMX_PINBASED_CTLS", MSR_IA32_VMX_PINBASED_CTLS, 1381 MSR_IA32_VMX_TRUE_PIN, 0x16 }, 1382 { "MSR_IA32_VMX_PROCBASED_CTLS", MSR_IA32_VMX_PROCBASED_CTLS, 1383 MSR_IA32_VMX_TRUE_PROC, 0x401e172 }, 1384 { "MSR_IA32_VMX_PROCBASED_CTLS2", MSR_IA32_VMX_PROCBASED_CTLS2, 1385 MSR_IA32_VMX_PROCBASED_CTLS2, 0 }, 1386 { "MSR_IA32_VMX_EXIT_CTLS", MSR_IA32_VMX_EXIT_CTLS, 1387 MSR_IA32_VMX_TRUE_EXIT, 0x36dff }, 1388 { "MSR_IA32_VMX_ENTRY_CTLS", MSR_IA32_VMX_ENTRY_CTLS, 1389 MSR_IA32_VMX_TRUE_ENTRY, 0x11ff }, 1390 }; 1391 1392 static void test_vmx_caps(void) 1393 { 1394 u64 val, default1, fixed0, fixed1; 1395 union vmx_ctrl_msr ctrl, true_ctrl; 1396 unsigned int n; 1397 bool ok; 1398 1399 printf("\nTest suite: VMX capability reporting\n"); 1400 1401 report("MSR_IA32_VMX_BASIC", 1402 (basic.revision & (1ul << 31)) == 0 && 1403 basic.size > 0 && basic.size <= 4096 && 1404 (basic.type == 0 || basic.type == 6) && 1405 basic.reserved1 == 0 && basic.reserved2 == 0); 1406 1407 val = rdmsr(MSR_IA32_VMX_MISC); 1408 report("MSR_IA32_VMX_MISC", 1409 (!(ctrl_cpu_rev[1].clr & CPU_URG) || val & (1ul << 5)) && 1410 ((val >> 16) & 0x1ff) <= 256 && 1411 (val & 0xc0007e00) == 0); 1412 1413 for (n = 0; n < ARRAY_SIZE(vmx_ctl_msr); n++) { 1414 ctrl.val = rdmsr(vmx_ctl_msr[n].index); 1415 default1 = vmx_ctl_msr[n].default1; 1416 ok = (ctrl.set & default1) == default1; 1417 ok = ok && (ctrl.set & ~ctrl.clr) == 0; 1418 if (ok && basic.ctrl) { 1419 true_ctrl.val = rdmsr(vmx_ctl_msr[n].true_index); 1420 ok = ctrl.clr == true_ctrl.clr; 1421 ok = ok && ctrl.set == (true_ctrl.set | default1); 1422 } 1423 report("%s", ok, vmx_ctl_msr[n].name); 1424 } 1425 1426 fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0); 1427 fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1); 1428 report("MSR_IA32_VMX_IA32_VMX_CR0_FIXED0/1", 1429 ((fixed0 ^ fixed1) & ~fixed1) == 0); 1430 1431 fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0); 1432 fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1); 1433 report("MSR_IA32_VMX_IA32_VMX_CR4_FIXED0/1", 1434 ((fixed0 ^ fixed1) & ~fixed1) == 0); 1435 1436 val = rdmsr(MSR_IA32_VMX_VMCS_ENUM); 1437 report("MSR_IA32_VMX_VMCS_ENUM", 1438 (val & 0x3e) >= 0x2a && 1439 (val & 0xfffffffffffffc01Ull) == 0); 1440 1441 val = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP); 1442 report("MSR_IA32_VMX_EPT_VPID_CAP", 1443 (val & 0xfffff07ef98cbebeUll) == 0); 1444 } 1445 1446 /* This function can only be called in guest */ 1447 static void __attribute__((__used__)) hypercall(u32 hypercall_no) 1448 { 1449 u64 val = 0; 1450 val = (hypercall_no & HYPERCALL_MASK) | HYPERCALL_BIT; 1451 hypercall_field = val; 1452 asm volatile("vmcall\n\t"); 1453 } 1454 1455 static bool is_hypercall() 1456 { 1457 ulong reason, hyper_bit; 1458 1459 reason = vmcs_read(EXI_REASON) & 0xff; 1460 hyper_bit = hypercall_field & HYPERCALL_BIT; 1461 if (reason == VMX_VMCALL && hyper_bit) 1462 return true; 1463 return false; 1464 } 1465 1466 static int handle_hypercall() 1467 { 1468 ulong hypercall_no; 1469 1470 hypercall_no = hypercall_field & HYPERCALL_MASK; 1471 hypercall_field = 0; 1472 switch (hypercall_no) { 1473 case HYPERCALL_VMEXIT: 1474 return VMX_TEST_VMEXIT; 1475 case HYPERCALL_VMABORT: 1476 return VMX_TEST_VMABORT; 1477 case HYPERCALL_VMSKIP: 1478 return VMX_TEST_VMSKIP; 1479 default: 1480 printf("ERROR : Invalid hypercall number : %ld\n", hypercall_no); 1481 } 1482 return VMX_TEST_EXIT; 1483 } 1484 1485 static void continue_abort(void) 1486 { 1487 assert(!in_guest); 1488 printf("Host was here when guest aborted:\n"); 1489 dump_stack(); 1490 longjmp(abort_target, 1); 1491 abort(); 1492 } 1493 1494 void __abort_test(void) 1495 { 1496 if (in_guest) 1497 hypercall(HYPERCALL_VMABORT); 1498 else 1499 longjmp(abort_target, 1); 1500 abort(); 1501 } 1502 1503 static void continue_skip(void) 1504 { 1505 assert(!in_guest); 1506 longjmp(abort_target, 1); 1507 abort(); 1508 } 1509 1510 void test_skip(const char *msg) 1511 { 1512 printf("%s skipping test: %s\n", in_guest ? "Guest" : "Host", msg); 1513 if (in_guest) 1514 hypercall(HYPERCALL_VMABORT); 1515 else 1516 longjmp(abort_target, 1); 1517 abort(); 1518 } 1519 1520 static int exit_handler() 1521 { 1522 int ret; 1523 1524 current->exits++; 1525 regs.rflags = vmcs_read(GUEST_RFLAGS); 1526 if (is_hypercall()) 1527 ret = handle_hypercall(); 1528 else 1529 ret = current->exit_handler(); 1530 vmcs_write(GUEST_RFLAGS, regs.rflags); 1531 1532 return ret; 1533 } 1534 1535 /* 1536 * Called if vmlaunch or vmresume fails. 1537 * @early - failure due to "VMX controls and host-state area" (26.2) 1538 * @vmlaunch - was this a vmlaunch or vmresume 1539 * @rflags - host rflags 1540 */ 1541 static int 1542 entry_failure_handler(struct vmentry_failure *failure) 1543 { 1544 if (current->entry_failure_handler) 1545 return current->entry_failure_handler(failure); 1546 else 1547 return VMX_TEST_EXIT; 1548 } 1549 1550 /* 1551 * Tries to enter the guest. Returns true iff entry succeeded. Otherwise, 1552 * populates @failure. 1553 */ 1554 static bool vmx_enter_guest(struct vmentry_failure *failure) 1555 { 1556 failure->early = 0; 1557 1558 in_guest = 1; 1559 asm volatile ( 1560 "mov %[HOST_RSP], %%rdi\n\t" 1561 "vmwrite %%rsp, %%rdi\n\t" 1562 LOAD_GPR_C 1563 "cmpb $0, %[launched]\n\t" 1564 "jne 1f\n\t" 1565 "vmlaunch\n\t" 1566 "jmp 2f\n\t" 1567 "1: " 1568 "vmresume\n\t" 1569 "2: " 1570 SAVE_GPR_C 1571 "pushf\n\t" 1572 "pop %%rdi\n\t" 1573 "mov %%rdi, %[failure_flags]\n\t" 1574 "movl $1, %[failure_flags]\n\t" 1575 "jmp 3f\n\t" 1576 "vmx_return:\n\t" 1577 SAVE_GPR_C 1578 "3: \n\t" 1579 : [failure_early]"+m"(failure->early), 1580 [failure_flags]"=m"(failure->flags) 1581 : [launched]"m"(launched), [HOST_RSP]"i"(HOST_RSP) 1582 : "rdi", "memory", "cc" 1583 ); 1584 in_guest = 0; 1585 1586 failure->vmlaunch = !launched; 1587 failure->instr = launched ? "vmresume" : "vmlaunch"; 1588 1589 return !failure->early && !(vmcs_read(EXI_REASON) & VMX_ENTRY_FAILURE); 1590 } 1591 1592 static int vmx_run() 1593 { 1594 while (1) { 1595 u32 ret; 1596 bool entered; 1597 struct vmentry_failure failure; 1598 1599 entered = vmx_enter_guest(&failure); 1600 1601 if (entered) { 1602 /* 1603 * VMCS isn't in "launched" state if there's been any 1604 * entry failure (early or otherwise). 1605 */ 1606 launched = 1; 1607 ret = exit_handler(); 1608 } else { 1609 ret = entry_failure_handler(&failure); 1610 } 1611 1612 switch (ret) { 1613 case VMX_TEST_RESUME: 1614 continue; 1615 case VMX_TEST_VMEXIT: 1616 guest_finished = 1; 1617 return 0; 1618 case VMX_TEST_EXIT: 1619 break; 1620 default: 1621 printf("ERROR : Invalid %s_handler return val %d.\n", 1622 entered ? "exit" : "entry_failure", 1623 ret); 1624 break; 1625 } 1626 1627 if (entered) 1628 print_vmexit_info(); 1629 else 1630 print_vmentry_failure_info(&failure); 1631 abort(); 1632 } 1633 } 1634 1635 static void run_teardown_step(struct test_teardown_step *step) 1636 { 1637 step->func(step->data); 1638 } 1639 1640 static int test_run(struct vmx_test *test) 1641 { 1642 int r; 1643 1644 /* Validate V2 interface. */ 1645 if (test->v2) { 1646 int ret = 0; 1647 if (test->init || test->guest_main || test->exit_handler || 1648 test->syscall_handler) { 1649 report("V2 test cannot specify V1 callbacks.", 0); 1650 ret = 1; 1651 } 1652 if (ret) 1653 return ret; 1654 } 1655 1656 if (test->name == NULL) 1657 test->name = "(no name)"; 1658 if (vmx_on()) { 1659 printf("%s : vmxon failed.\n", __func__); 1660 return 1; 1661 } 1662 1663 init_vmcs(&(test->vmcs)); 1664 /* Directly call test->init is ok here, init_vmcs has done 1665 vmcs init, vmclear and vmptrld*/ 1666 if (test->init && test->init(test->vmcs) != VMX_TEST_START) 1667 goto out; 1668 teardown_count = 0; 1669 v2_guest_main = NULL; 1670 test->exits = 0; 1671 current = test; 1672 regs = test->guest_regs; 1673 vmcs_write(GUEST_RFLAGS, regs.rflags | 0x2); 1674 launched = 0; 1675 guest_finished = 0; 1676 printf("\nTest suite: %s\n", test->name); 1677 1678 r = setjmp(abort_target); 1679 if (r) { 1680 assert(!in_guest); 1681 goto out; 1682 } 1683 1684 1685 if (test->v2) 1686 test->v2(); 1687 else 1688 vmx_run(); 1689 1690 while (teardown_count > 0) 1691 run_teardown_step(&teardown_steps[--teardown_count]); 1692 1693 if (launched && !guest_finished) 1694 report("Guest didn't run to completion.", 0); 1695 1696 out: 1697 if (vmx_off()) { 1698 printf("%s : vmxoff failed.\n", __func__); 1699 return 1; 1700 } 1701 return 0; 1702 } 1703 1704 /* 1705 * Add a teardown step. Executed after the test's main function returns. 1706 * Teardown steps executed in reverse order. 1707 */ 1708 void test_add_teardown(test_teardown_func func, void *data) 1709 { 1710 struct test_teardown_step *step; 1711 1712 TEST_ASSERT_MSG(teardown_count < MAX_TEST_TEARDOWN_STEPS, 1713 "There are already %d teardown steps.", 1714 teardown_count); 1715 step = &teardown_steps[teardown_count++]; 1716 step->func = func; 1717 step->data = data; 1718 } 1719 1720 /* 1721 * Set the target of the first enter_guest call. Can only be called once per 1722 * test. Must be called before first enter_guest call. 1723 */ 1724 void test_set_guest(test_guest_func func) 1725 { 1726 assert(current->v2); 1727 TEST_ASSERT_MSG(!v2_guest_main, "Already set guest func."); 1728 v2_guest_main = func; 1729 } 1730 1731 /* 1732 * Enters the guest (or launches it for the first time). Error to call once the 1733 * guest has returned (i.e., run past the end of its guest() function). Also 1734 * aborts if guest entry fails. 1735 */ 1736 void enter_guest(void) 1737 { 1738 struct vmentry_failure failure; 1739 1740 TEST_ASSERT_MSG(v2_guest_main, 1741 "Never called test_set_guest_func!"); 1742 1743 TEST_ASSERT_MSG(!guest_finished, 1744 "Called enter_guest() after guest returned."); 1745 1746 if (!vmx_enter_guest(&failure)) { 1747 print_vmentry_failure_info(&failure); 1748 abort(); 1749 } 1750 1751 launched = 1; 1752 1753 if (is_hypercall()) { 1754 int ret; 1755 1756 ret = handle_hypercall(); 1757 switch (ret) { 1758 case VMX_TEST_VMEXIT: 1759 guest_finished = 1; 1760 break; 1761 case VMX_TEST_VMABORT: 1762 continue_abort(); 1763 break; 1764 case VMX_TEST_VMSKIP: 1765 continue_skip(); 1766 break; 1767 default: 1768 printf("ERROR : Invalid handle_hypercall return %d.\n", 1769 ret); 1770 abort(); 1771 } 1772 } 1773 } 1774 1775 extern struct vmx_test vmx_tests[]; 1776 1777 static bool 1778 test_wanted(const char *name, const char *filters[], int filter_count) 1779 { 1780 int i; 1781 bool positive = false; 1782 bool match = false; 1783 char clean_name[strlen(name) + 1]; 1784 char *c; 1785 const char *n; 1786 1787 /* Replace spaces with underscores. */ 1788 n = name; 1789 c = &clean_name[0]; 1790 do *c++ = (*n == ' ') ? '_' : *n; 1791 while (*n++); 1792 1793 for (i = 0; i < filter_count; i++) { 1794 const char *filter = filters[i]; 1795 1796 if (filter[0] == '-') { 1797 if (simple_glob(clean_name, filter + 1)) 1798 return false; 1799 } else { 1800 positive = true; 1801 match |= simple_glob(clean_name, filter); 1802 } 1803 } 1804 1805 if (!positive || match) { 1806 matched++; 1807 return true; 1808 } else { 1809 return false; 1810 } 1811 } 1812 1813 int main(int argc, const char *argv[]) 1814 { 1815 int i = 0; 1816 1817 setup_vm(); 1818 setup_idt(); 1819 hypercall_field = 0; 1820 1821 argv++; 1822 argc--; 1823 1824 if (!(cpuid(1).c & (1 << 5))) { 1825 printf("WARNING: vmx not supported, add '-cpu host'\n"); 1826 goto exit; 1827 } 1828 init_vmx(); 1829 if (test_wanted("test_vmx_feature_control", argv, argc)) { 1830 /* Sets MSR_IA32_FEATURE_CONTROL to 0x5 */ 1831 if (test_vmx_feature_control() != 0) 1832 goto exit; 1833 } else { 1834 if ((rdmsr(MSR_IA32_FEATURE_CONTROL) & 0x5) != 0x5) 1835 wrmsr(MSR_IA32_FEATURE_CONTROL, 0x5); 1836 } 1837 1838 if (test_wanted("test_vmxon", argv, argc)) { 1839 /* Enables VMX */ 1840 if (test_vmxon() != 0) 1841 goto exit; 1842 } else { 1843 if (vmx_on()) { 1844 report("vmxon", 0); 1845 goto exit; 1846 } 1847 } 1848 1849 if (test_wanted("test_vmptrld", argv, argc)) 1850 test_vmptrld(); 1851 if (test_wanted("test_vmclear", argv, argc)) 1852 test_vmclear(); 1853 if (test_wanted("test_vmptrst", argv, argc)) 1854 test_vmptrst(); 1855 if (test_wanted("test_vmwrite_vmread", argv, argc)) 1856 test_vmwrite_vmread(); 1857 if (test_wanted("test_vmcs_high", argv, argc)) 1858 test_vmcs_high(); 1859 if (test_wanted("test_vmcs_lifecycle", argv, argc)) 1860 test_vmcs_lifecycle(); 1861 if (test_wanted("test_vmx_caps", argv, argc)) 1862 test_vmx_caps(); 1863 1864 /* Balance vmxon from test_vmxon. */ 1865 vmx_off(); 1866 1867 for (; vmx_tests[i].name != NULL; i++) { 1868 if (!test_wanted(vmx_tests[i].name, argv, argc)) 1869 continue; 1870 if (test_run(&vmx_tests[i])) 1871 goto exit; 1872 } 1873 1874 if (!matched) 1875 report("command line didn't match any tests!", matched); 1876 1877 exit: 1878 return report_summary(); 1879 } 1880