1 /* 2 * All test cases of nested virtualization should be in this file 3 * 4 * Author : Arthur Chunqi Li <yzt356@gmail.com> 5 */ 6 7 #include <asm/debugreg.h> 8 9 #include "vmx.h" 10 #include "msr.h" 11 #include "processor.h" 12 #include "vm.h" 13 #include "pci.h" 14 #include "fwcfg.h" 15 #include "isr.h" 16 #include "desc.h" 17 #include "apic.h" 18 #include "types.h" 19 #include "vmalloc.h" 20 #include "alloc_page.h" 21 #include "smp.h" 22 #include "delay.h" 23 24 #define NONCANONICAL 0xaaaaaaaaaaaaaaaaull 25 26 #define VPID_CAP_INVVPID_TYPES_SHIFT 40 27 28 u64 ia32_pat; 29 u64 ia32_efer; 30 void *io_bitmap_a, *io_bitmap_b; 31 u16 ioport; 32 33 unsigned long *pml4; 34 u64 eptp; 35 void *data_page1, *data_page2; 36 37 phys_addr_t pci_physaddr; 38 39 void *pml_log; 40 #define PML_INDEX 512 41 42 static inline unsigned ffs(unsigned x) 43 { 44 int pos = -1; 45 46 __asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0" 47 : "+r"(pos) : "rm"(x) : "eax"); 48 return pos + 1; 49 } 50 51 static inline void vmcall(void) 52 { 53 asm volatile("vmcall"); 54 } 55 56 static void basic_guest_main(void) 57 { 58 report(1, "Basic VMX test"); 59 } 60 61 static int basic_exit_handler(void) 62 { 63 report(0, "Basic VMX test"); 64 print_vmexit_info(); 65 return VMX_TEST_EXIT; 66 } 67 68 static void vmenter_main(void) 69 { 70 u64 rax; 71 u64 rsp, resume_rsp; 72 73 report(1, "test vmlaunch"); 74 75 asm volatile( 76 "mov %%rsp, %0\n\t" 77 "mov %3, %%rax\n\t" 78 "vmcall\n\t" 79 "mov %%rax, %1\n\t" 80 "mov %%rsp, %2\n\t" 81 : "=r"(rsp), "=r"(rax), "=r"(resume_rsp) 82 : "g"(0xABCD)); 83 report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume"); 84 } 85 86 static int vmenter_exit_handler(void) 87 { 88 u64 guest_rip; 89 ulong reason; 90 91 guest_rip = vmcs_read(GUEST_RIP); 92 reason = vmcs_read(EXI_REASON) & 0xff; 93 switch (reason) { 94 case VMX_VMCALL: 95 if (regs.rax != 0xABCD) { 96 report(0, "test vmresume"); 97 return VMX_TEST_VMEXIT; 98 } 99 regs.rax = 0xFFFF; 100 vmcs_write(GUEST_RIP, guest_rip + 3); 101 return VMX_TEST_RESUME; 102 default: 103 report(0, "test vmresume"); 104 print_vmexit_info(); 105 } 106 return VMX_TEST_VMEXIT; 107 } 108 109 u32 preempt_scale; 110 volatile unsigned long long tsc_val; 111 volatile u32 preempt_val; 112 u64 saved_rip; 113 114 static int preemption_timer_init(struct vmcs *vmcs) 115 { 116 if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) { 117 printf("\tPreemption timer is not supported\n"); 118 return VMX_TEST_EXIT; 119 } 120 vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT); 121 preempt_val = 10000000; 122 vmcs_write(PREEMPT_TIMER_VALUE, preempt_val); 123 preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F; 124 125 if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT)) 126 printf("\tSave preemption value is not supported\n"); 127 128 return VMX_TEST_START; 129 } 130 131 static void preemption_timer_main(void) 132 { 133 tsc_val = rdtsc(); 134 if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) { 135 vmx_set_test_stage(0); 136 vmcall(); 137 if (vmx_get_test_stage() == 1) 138 vmcall(); 139 } 140 vmx_set_test_stage(1); 141 while (vmx_get_test_stage() == 1) { 142 if (((rdtsc() - tsc_val) >> preempt_scale) 143 > 10 * preempt_val) { 144 vmx_set_test_stage(2); 145 vmcall(); 146 } 147 } 148 tsc_val = rdtsc(); 149 asm volatile ("hlt"); 150 vmcall(); 151 vmx_set_test_stage(5); 152 vmcall(); 153 } 154 155 static int preemption_timer_exit_handler(void) 156 { 157 bool guest_halted; 158 u64 guest_rip; 159 ulong reason; 160 u32 insn_len; 161 u32 ctrl_exit; 162 163 guest_rip = vmcs_read(GUEST_RIP); 164 reason = vmcs_read(EXI_REASON) & 0xff; 165 insn_len = vmcs_read(EXI_INST_LEN); 166 switch (reason) { 167 case VMX_PREEMPT: 168 switch (vmx_get_test_stage()) { 169 case 1: 170 case 2: 171 report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val, 172 "busy-wait for preemption timer"); 173 vmx_set_test_stage(3); 174 vmcs_write(PREEMPT_TIMER_VALUE, preempt_val); 175 return VMX_TEST_RESUME; 176 case 3: 177 guest_halted = 178 (vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT); 179 report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val 180 && guest_halted, 181 "preemption timer during hlt"); 182 vmx_set_test_stage(4); 183 vmcs_write(PIN_CONTROLS, 184 vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT); 185 vmcs_write(EXI_CONTROLS, 186 vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT); 187 vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE); 188 return VMX_TEST_RESUME; 189 case 4: 190 report(saved_rip == guest_rip, 191 "preemption timer with 0 value"); 192 break; 193 default: 194 report(false, "Invalid stage."); 195 print_vmexit_info(); 196 break; 197 } 198 break; 199 case VMX_VMCALL: 200 vmcs_write(GUEST_RIP, guest_rip + insn_len); 201 switch (vmx_get_test_stage()) { 202 case 0: 203 report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val, 204 "Keep preemption value"); 205 vmx_set_test_stage(1); 206 vmcs_write(PREEMPT_TIMER_VALUE, preempt_val); 207 ctrl_exit = (vmcs_read(EXI_CONTROLS) | 208 EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr; 209 vmcs_write(EXI_CONTROLS, ctrl_exit); 210 return VMX_TEST_RESUME; 211 case 1: 212 report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val, 213 "Save preemption value"); 214 return VMX_TEST_RESUME; 215 case 2: 216 report(0, "busy-wait for preemption timer"); 217 vmx_set_test_stage(3); 218 vmcs_write(PREEMPT_TIMER_VALUE, preempt_val); 219 return VMX_TEST_RESUME; 220 case 3: 221 report(0, "preemption timer during hlt"); 222 vmx_set_test_stage(4); 223 /* fall through */ 224 case 4: 225 vmcs_write(PIN_CONTROLS, 226 vmcs_read(PIN_CONTROLS) | PIN_PREEMPT); 227 vmcs_write(PREEMPT_TIMER_VALUE, 0); 228 saved_rip = guest_rip + insn_len; 229 return VMX_TEST_RESUME; 230 case 5: 231 report(0, 232 "preemption timer with 0 value (vmcall stage 5)"); 233 break; 234 default: 235 // Should not reach here 236 report(false, "unexpected stage, %d", 237 vmx_get_test_stage()); 238 print_vmexit_info(); 239 return VMX_TEST_VMEXIT; 240 } 241 break; 242 default: 243 report(false, "Unknown exit reason, %ld", reason); 244 print_vmexit_info(); 245 } 246 vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT); 247 return VMX_TEST_VMEXIT; 248 } 249 250 static void msr_bmp_init(void) 251 { 252 void *msr_bitmap; 253 u32 ctrl_cpu0; 254 255 msr_bitmap = alloc_page(); 256 ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0); 257 ctrl_cpu0 |= CPU_MSR_BITMAP; 258 vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0); 259 vmcs_write(MSR_BITMAP, (u64)msr_bitmap); 260 } 261 262 static void *get_msr_bitmap(void) 263 { 264 void *msr_bitmap; 265 266 if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) { 267 msr_bitmap = (void *)vmcs_read(MSR_BITMAP); 268 } else { 269 msr_bitmap = alloc_page(); 270 memset(msr_bitmap, 0xff, PAGE_SIZE); 271 vmcs_write(MSR_BITMAP, (u64)msr_bitmap); 272 vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP); 273 } 274 275 return msr_bitmap; 276 } 277 278 static void disable_intercept_for_x2apic_msrs(void) 279 { 280 unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap(); 281 u32 msr; 282 283 for (msr = APIC_BASE_MSR; 284 msr < (APIC_BASE_MSR+0xff); 285 msr += BITS_PER_LONG) { 286 unsigned int word = msr / BITS_PER_LONG; 287 288 msr_bitmap[word] = 0; 289 msr_bitmap[word + (0x800 / sizeof(long))] = 0; 290 } 291 } 292 293 static int test_ctrl_pat_init(struct vmcs *vmcs) 294 { 295 u64 ctrl_ent; 296 u64 ctrl_exi; 297 298 msr_bmp_init(); 299 if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) && 300 !(ctrl_exit_rev.clr & EXI_LOAD_PAT) && 301 !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) { 302 printf("\tSave/load PAT is not supported\n"); 303 return 1; 304 } 305 306 ctrl_ent = vmcs_read(ENT_CONTROLS); 307 ctrl_exi = vmcs_read(EXI_CONTROLS); 308 ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT; 309 ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT); 310 vmcs_write(ENT_CONTROLS, ctrl_ent); 311 vmcs_write(EXI_CONTROLS, ctrl_exi); 312 ia32_pat = rdmsr(MSR_IA32_CR_PAT); 313 vmcs_write(GUEST_PAT, 0x0); 314 vmcs_write(HOST_PAT, ia32_pat); 315 return VMX_TEST_START; 316 } 317 318 static void test_ctrl_pat_main(void) 319 { 320 u64 guest_ia32_pat; 321 322 guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT); 323 if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) 324 printf("\tENT_LOAD_PAT is not supported.\n"); 325 else { 326 if (guest_ia32_pat != 0) { 327 report(0, "Entry load PAT"); 328 return; 329 } 330 } 331 wrmsr(MSR_IA32_CR_PAT, 0x6); 332 vmcall(); 333 guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT); 334 if (ctrl_enter_rev.clr & ENT_LOAD_PAT) 335 report(guest_ia32_pat == ia32_pat, "Entry load PAT"); 336 } 337 338 static int test_ctrl_pat_exit_handler(void) 339 { 340 u64 guest_rip; 341 ulong reason; 342 u64 guest_pat; 343 344 guest_rip = vmcs_read(GUEST_RIP); 345 reason = vmcs_read(EXI_REASON) & 0xff; 346 switch (reason) { 347 case VMX_VMCALL: 348 guest_pat = vmcs_read(GUEST_PAT); 349 if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) { 350 printf("\tEXI_SAVE_PAT is not supported\n"); 351 vmcs_write(GUEST_PAT, 0x6); 352 } else { 353 report(guest_pat == 0x6, "Exit save PAT"); 354 } 355 if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) 356 printf("\tEXI_LOAD_PAT is not supported\n"); 357 else 358 report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat, 359 "Exit load PAT"); 360 vmcs_write(GUEST_PAT, ia32_pat); 361 vmcs_write(GUEST_RIP, guest_rip + 3); 362 return VMX_TEST_RESUME; 363 default: 364 printf("ERROR : Undefined exit reason, reason = %ld.\n", reason); 365 break; 366 } 367 return VMX_TEST_VMEXIT; 368 } 369 370 static int test_ctrl_efer_init(struct vmcs *vmcs) 371 { 372 u64 ctrl_ent; 373 u64 ctrl_exi; 374 375 msr_bmp_init(); 376 ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER; 377 ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER; 378 vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr); 379 vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr); 380 ia32_efer = rdmsr(MSR_EFER); 381 vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX); 382 vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX); 383 return VMX_TEST_START; 384 } 385 386 static void test_ctrl_efer_main(void) 387 { 388 u64 guest_ia32_efer; 389 390 guest_ia32_efer = rdmsr(MSR_EFER); 391 if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) 392 printf("\tENT_LOAD_EFER is not supported.\n"); 393 else { 394 if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) { 395 report(0, "Entry load EFER"); 396 return; 397 } 398 } 399 wrmsr(MSR_EFER, ia32_efer); 400 vmcall(); 401 guest_ia32_efer = rdmsr(MSR_EFER); 402 if (ctrl_enter_rev.clr & ENT_LOAD_EFER) 403 report(guest_ia32_efer == ia32_efer, "Entry load EFER"); 404 } 405 406 static int test_ctrl_efer_exit_handler(void) 407 { 408 u64 guest_rip; 409 ulong reason; 410 u64 guest_efer; 411 412 guest_rip = vmcs_read(GUEST_RIP); 413 reason = vmcs_read(EXI_REASON) & 0xff; 414 switch (reason) { 415 case VMX_VMCALL: 416 guest_efer = vmcs_read(GUEST_EFER); 417 if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) { 418 printf("\tEXI_SAVE_EFER is not supported\n"); 419 vmcs_write(GUEST_EFER, ia32_efer); 420 } else { 421 report(guest_efer == ia32_efer, "Exit save EFER"); 422 } 423 if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) { 424 printf("\tEXI_LOAD_EFER is not supported\n"); 425 wrmsr(MSR_EFER, ia32_efer ^ EFER_NX); 426 } else { 427 report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX), 428 "Exit load EFER"); 429 } 430 vmcs_write(GUEST_PAT, ia32_efer); 431 vmcs_write(GUEST_RIP, guest_rip + 3); 432 return VMX_TEST_RESUME; 433 default: 434 printf("ERROR : Undefined exit reason, reason = %ld.\n", reason); 435 break; 436 } 437 return VMX_TEST_VMEXIT; 438 } 439 440 u32 guest_cr0, guest_cr4; 441 442 static void cr_shadowing_main(void) 443 { 444 u32 cr0, cr4, tmp; 445 446 // Test read through 447 vmx_set_test_stage(0); 448 guest_cr0 = read_cr0(); 449 if (vmx_get_test_stage() == 1) 450 report(0, "Read through CR0"); 451 else 452 vmcall(); 453 vmx_set_test_stage(1); 454 guest_cr4 = read_cr4(); 455 if (vmx_get_test_stage() == 2) 456 report(0, "Read through CR4"); 457 else 458 vmcall(); 459 // Test write through 460 guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP); 461 guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE); 462 vmx_set_test_stage(2); 463 write_cr0(guest_cr0); 464 if (vmx_get_test_stage() == 3) 465 report(0, "Write throuth CR0"); 466 else 467 vmcall(); 468 vmx_set_test_stage(3); 469 write_cr4(guest_cr4); 470 if (vmx_get_test_stage() == 4) 471 report(0, "Write through CR4"); 472 else 473 vmcall(); 474 // Test read shadow 475 vmx_set_test_stage(4); 476 vmcall(); 477 cr0 = read_cr0(); 478 if (vmx_get_test_stage() != 5) 479 report(cr0 == guest_cr0, "Read shadowing CR0"); 480 vmx_set_test_stage(5); 481 cr4 = read_cr4(); 482 if (vmx_get_test_stage() != 6) 483 report(cr4 == guest_cr4, "Read shadowing CR4"); 484 // Test write shadow (same value with shadow) 485 vmx_set_test_stage(6); 486 write_cr0(guest_cr0); 487 if (vmx_get_test_stage() == 7) 488 report(0, "Write shadowing CR0 (same value with shadow)"); 489 else 490 vmcall(); 491 vmx_set_test_stage(7); 492 write_cr4(guest_cr4); 493 if (vmx_get_test_stage() == 8) 494 report(0, "Write shadowing CR4 (same value with shadow)"); 495 else 496 vmcall(); 497 // Test write shadow (different value) 498 vmx_set_test_stage(8); 499 tmp = guest_cr0 ^ X86_CR0_TS; 500 asm volatile("mov %0, %%rsi\n\t" 501 "mov %%rsi, %%cr0\n\t" 502 ::"m"(tmp) 503 :"rsi", "memory", "cc"); 504 report(vmx_get_test_stage() == 9, 505 "Write shadowing different X86_CR0_TS"); 506 vmx_set_test_stage(9); 507 tmp = guest_cr0 ^ X86_CR0_MP; 508 asm volatile("mov %0, %%rsi\n\t" 509 "mov %%rsi, %%cr0\n\t" 510 ::"m"(tmp) 511 :"rsi", "memory", "cc"); 512 report(vmx_get_test_stage() == 10, 513 "Write shadowing different X86_CR0_MP"); 514 vmx_set_test_stage(10); 515 tmp = guest_cr4 ^ X86_CR4_TSD; 516 asm volatile("mov %0, %%rsi\n\t" 517 "mov %%rsi, %%cr4\n\t" 518 ::"m"(tmp) 519 :"rsi", "memory", "cc"); 520 report(vmx_get_test_stage() == 11, 521 "Write shadowing different X86_CR4_TSD"); 522 vmx_set_test_stage(11); 523 tmp = guest_cr4 ^ X86_CR4_DE; 524 asm volatile("mov %0, %%rsi\n\t" 525 "mov %%rsi, %%cr4\n\t" 526 ::"m"(tmp) 527 :"rsi", "memory", "cc"); 528 report(vmx_get_test_stage() == 12, 529 "Write shadowing different X86_CR4_DE"); 530 } 531 532 static int cr_shadowing_exit_handler(void) 533 { 534 u64 guest_rip; 535 ulong reason; 536 u32 insn_len; 537 u32 exit_qual; 538 539 guest_rip = vmcs_read(GUEST_RIP); 540 reason = vmcs_read(EXI_REASON) & 0xff; 541 insn_len = vmcs_read(EXI_INST_LEN); 542 exit_qual = vmcs_read(EXI_QUALIFICATION); 543 switch (reason) { 544 case VMX_VMCALL: 545 switch (vmx_get_test_stage()) { 546 case 0: 547 report(guest_cr0 == vmcs_read(GUEST_CR0), 548 "Read through CR0"); 549 break; 550 case 1: 551 report(guest_cr4 == vmcs_read(GUEST_CR4), 552 "Read through CR4"); 553 break; 554 case 2: 555 report(guest_cr0 == vmcs_read(GUEST_CR0), 556 "Write through CR0"); 557 break; 558 case 3: 559 report(guest_cr4 == vmcs_read(GUEST_CR4), 560 "Write through CR4"); 561 break; 562 case 4: 563 guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP); 564 guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE); 565 vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP); 566 vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP)); 567 vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE); 568 vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE)); 569 break; 570 case 6: 571 report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)), 572 "Write shadowing CR0 (same value)"); 573 break; 574 case 7: 575 report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)), 576 "Write shadowing CR4 (same value)"); 577 break; 578 default: 579 // Should not reach here 580 report(false, "unexpected stage, %d", 581 vmx_get_test_stage()); 582 print_vmexit_info(); 583 return VMX_TEST_VMEXIT; 584 } 585 vmcs_write(GUEST_RIP, guest_rip + insn_len); 586 return VMX_TEST_RESUME; 587 case VMX_CR: 588 switch (vmx_get_test_stage()) { 589 case 4: 590 report(0, "Read shadowing CR0"); 591 vmx_inc_test_stage(); 592 break; 593 case 5: 594 report(0, "Read shadowing CR4"); 595 vmx_inc_test_stage(); 596 break; 597 case 6: 598 report(0, "Write shadowing CR0 (same value)"); 599 vmx_inc_test_stage(); 600 break; 601 case 7: 602 report(0, "Write shadowing CR4 (same value)"); 603 vmx_inc_test_stage(); 604 break; 605 case 8: 606 case 9: 607 // 0x600 encodes "mov %esi, %cr0" 608 if (exit_qual == 0x600) 609 vmx_inc_test_stage(); 610 break; 611 case 10: 612 case 11: 613 // 0x604 encodes "mov %esi, %cr4" 614 if (exit_qual == 0x604) 615 vmx_inc_test_stage(); 616 break; 617 default: 618 // Should not reach here 619 report(false, "unexpected stage, %d", 620 vmx_get_test_stage()); 621 print_vmexit_info(); 622 return VMX_TEST_VMEXIT; 623 } 624 vmcs_write(GUEST_RIP, guest_rip + insn_len); 625 return VMX_TEST_RESUME; 626 default: 627 report(false, "Unknown exit reason, %ld", reason); 628 print_vmexit_info(); 629 } 630 return VMX_TEST_VMEXIT; 631 } 632 633 static int iobmp_init(struct vmcs *vmcs) 634 { 635 u32 ctrl_cpu0; 636 637 io_bitmap_a = alloc_page(); 638 io_bitmap_b = alloc_page(); 639 ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0); 640 ctrl_cpu0 |= CPU_IO_BITMAP; 641 ctrl_cpu0 &= (~CPU_IO); 642 vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0); 643 vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a); 644 vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b); 645 return VMX_TEST_START; 646 } 647 648 static void iobmp_main(void) 649 { 650 // stage 0, test IO pass 651 vmx_set_test_stage(0); 652 inb(0x5000); 653 outb(0x0, 0x5000); 654 report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass"); 655 // test IO width, in/out 656 ((u8 *)io_bitmap_a)[0] = 0xFF; 657 vmx_set_test_stage(2); 658 inb(0x0); 659 report(vmx_get_test_stage() == 3, "I/O bitmap - trap in"); 660 vmx_set_test_stage(3); 661 outw(0x0, 0x0); 662 report(vmx_get_test_stage() == 4, "I/O bitmap - trap out"); 663 vmx_set_test_stage(4); 664 inl(0x0); 665 report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long"); 666 // test low/high IO port 667 vmx_set_test_stage(5); 668 ((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8)); 669 inb(0x5000); 670 report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part"); 671 vmx_set_test_stage(6); 672 ((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8)); 673 inb(0x9000); 674 report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part"); 675 // test partial pass 676 vmx_set_test_stage(7); 677 inl(0x4FFF); 678 report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass"); 679 // test overrun 680 vmx_set_test_stage(8); 681 memset(io_bitmap_a, 0x0, PAGE_SIZE); 682 memset(io_bitmap_b, 0x0, PAGE_SIZE); 683 inl(0xFFFF); 684 report(vmx_get_test_stage() == 9, "I/O bitmap - overrun"); 685 vmx_set_test_stage(9); 686 vmcall(); 687 outb(0x0, 0x0); 688 report(vmx_get_test_stage() == 9, 689 "I/O bitmap - ignore unconditional exiting"); 690 vmx_set_test_stage(10); 691 vmcall(); 692 outb(0x0, 0x0); 693 report(vmx_get_test_stage() == 11, 694 "I/O bitmap - unconditional exiting"); 695 } 696 697 static int iobmp_exit_handler(void) 698 { 699 u64 guest_rip; 700 ulong reason, exit_qual; 701 u32 insn_len, ctrl_cpu0; 702 703 guest_rip = vmcs_read(GUEST_RIP); 704 reason = vmcs_read(EXI_REASON) & 0xff; 705 exit_qual = vmcs_read(EXI_QUALIFICATION); 706 insn_len = vmcs_read(EXI_INST_LEN); 707 switch (reason) { 708 case VMX_IO: 709 switch (vmx_get_test_stage()) { 710 case 0: 711 case 1: 712 vmx_inc_test_stage(); 713 break; 714 case 2: 715 report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE, 716 "I/O bitmap - I/O width, byte"); 717 report(exit_qual & VMX_IO_IN, 718 "I/O bitmap - I/O direction, in"); 719 vmx_inc_test_stage(); 720 break; 721 case 3: 722 report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD, 723 "I/O bitmap - I/O width, word"); 724 report(!(exit_qual & VMX_IO_IN), 725 "I/O bitmap - I/O direction, out"); 726 vmx_inc_test_stage(); 727 break; 728 case 4: 729 report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG, 730 "I/O bitmap - I/O width, long"); 731 vmx_inc_test_stage(); 732 break; 733 case 5: 734 if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000) 735 vmx_inc_test_stage(); 736 break; 737 case 6: 738 if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000) 739 vmx_inc_test_stage(); 740 break; 741 case 7: 742 if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF) 743 vmx_inc_test_stage(); 744 break; 745 case 8: 746 if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF) 747 vmx_inc_test_stage(); 748 break; 749 case 9: 750 case 10: 751 ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0); 752 vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO); 753 vmx_inc_test_stage(); 754 break; 755 default: 756 // Should not reach here 757 report(false, "unexpected stage, %d", 758 vmx_get_test_stage()); 759 print_vmexit_info(); 760 return VMX_TEST_VMEXIT; 761 } 762 vmcs_write(GUEST_RIP, guest_rip + insn_len); 763 return VMX_TEST_RESUME; 764 case VMX_VMCALL: 765 switch (vmx_get_test_stage()) { 766 case 9: 767 ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0); 768 ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP; 769 vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0); 770 break; 771 case 10: 772 ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0); 773 ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO; 774 vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0); 775 break; 776 default: 777 // Should not reach here 778 report(false, "unexpected stage, %d", 779 vmx_get_test_stage()); 780 print_vmexit_info(); 781 return VMX_TEST_VMEXIT; 782 } 783 vmcs_write(GUEST_RIP, guest_rip + insn_len); 784 return VMX_TEST_RESUME; 785 default: 786 printf("guest_rip = %#lx\n", guest_rip); 787 printf("\tERROR : Undefined exit reason, reason = %ld.\n", reason); 788 break; 789 } 790 return VMX_TEST_VMEXIT; 791 } 792 793 #define INSN_CPU0 0 794 #define INSN_CPU1 1 795 #define INSN_ALWAYS_TRAP 2 796 797 #define FIELD_EXIT_QUAL (1 << 0) 798 #define FIELD_INSN_INFO (1 << 1) 799 800 asm( 801 "insn_hlt: hlt;ret\n\t" 802 "insn_invlpg: invlpg 0x12345678;ret\n\t" 803 "insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t" 804 "insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t" 805 "insn_rdtsc: rdtsc;ret\n\t" 806 "insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t" 807 "insn_cr3_store: mov %cr3,%rax;ret\n\t" 808 #ifdef __x86_64__ 809 "insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t" 810 "insn_cr8_store: mov %cr8,%rax;ret\n\t" 811 #endif 812 "insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t" 813 "insn_pause: pause;ret\n\t" 814 "insn_wbinvd: wbinvd;ret\n\t" 815 "insn_cpuid: mov $10, %eax; cpuid;ret\n\t" 816 "insn_invd: invd;ret\n\t" 817 "insn_sgdt: sgdt gdt64_desc;ret\n\t" 818 "insn_lgdt: lgdt gdt64_desc;ret\n\t" 819 "insn_sidt: sidt idt_descr;ret\n\t" 820 "insn_lidt: lidt idt_descr;ret\n\t" 821 "insn_sldt: sldt %ax;ret\n\t" 822 "insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t" 823 "insn_str: str %ax;ret\n\t" 824 "insn_rdrand: rdrand %rax;ret\n\t" 825 "insn_rdseed: rdseed %rax;ret\n\t" 826 ); 827 extern void insn_hlt(void); 828 extern void insn_invlpg(void); 829 extern void insn_mwait(void); 830 extern void insn_rdpmc(void); 831 extern void insn_rdtsc(void); 832 extern void insn_cr3_load(void); 833 extern void insn_cr3_store(void); 834 #ifdef __x86_64__ 835 extern void insn_cr8_load(void); 836 extern void insn_cr8_store(void); 837 #endif 838 extern void insn_monitor(void); 839 extern void insn_pause(void); 840 extern void insn_wbinvd(void); 841 extern void insn_sgdt(void); 842 extern void insn_lgdt(void); 843 extern void insn_sidt(void); 844 extern void insn_lidt(void); 845 extern void insn_sldt(void); 846 extern void insn_lldt(void); 847 extern void insn_str(void); 848 extern void insn_cpuid(void); 849 extern void insn_invd(void); 850 extern void insn_rdrand(void); 851 extern void insn_rdseed(void); 852 853 u32 cur_insn; 854 u64 cr3; 855 856 #define X86_FEATURE_MONITOR (1 << 3) 857 858 typedef bool (*supported_fn)(void); 859 860 static bool monitor_supported(void) 861 { 862 return this_cpu_has(X86_FEATURE_MWAIT); 863 } 864 865 struct insn_table { 866 const char *name; 867 u32 flag; 868 void (*insn_func)(void); 869 u32 type; 870 u32 reason; 871 ulong exit_qual; 872 u32 insn_info; 873 // Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define 874 // which field need to be tested, reason is always tested 875 u32 test_field; 876 const supported_fn supported_fn; 877 u8 disabled; 878 }; 879 880 /* 881 * Add more test cases of instruction intercept here. Elements in this 882 * table is: 883 * name/control flag/insn function/type/exit reason/exit qulification/ 884 * instruction info/field to test 885 * The last field defines which fields (exit_qual and insn_info) need to be 886 * tested in exit handler. If set to 0, only "reason" is checked. 887 */ 888 static struct insn_table insn_table[] = { 889 // Flags for Primary Processor-Based VM-Execution Controls 890 {"HLT", CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0}, 891 {"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14, 892 0x12345678, 0, FIELD_EXIT_QUAL}, 893 {"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported}, 894 {"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0}, 895 {"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0}, 896 {"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0, 897 FIELD_EXIT_QUAL}, 898 {"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0, 899 FIELD_EXIT_QUAL}, 900 #ifdef __x86_64__ 901 {"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0, 902 FIELD_EXIT_QUAL}, 903 {"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0, 904 FIELD_EXIT_QUAL}, 905 #endif 906 {"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported}, 907 {"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0}, 908 // Flags for Secondary Processor-Based VM-Execution Controls 909 {"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0}, 910 {"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0}, 911 {"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0}, 912 {"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0}, 913 {"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0}, 914 {"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0}, 915 {"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0}, 916 {"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0}, 917 /* LTR causes a #GP if done with a busy selector, so it is not tested. */ 918 {"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0}, 919 {"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0}, 920 // Instructions always trap 921 {"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0}, 922 {"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0}, 923 // Instructions never trap 924 {NULL}, 925 }; 926 927 static int insn_intercept_init(struct vmcs *vmcs) 928 { 929 u32 ctrl_cpu, cur_insn; 930 931 ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY; 932 ctrl_cpu &= ctrl_cpu_rev[0].clr; 933 vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu); 934 vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set); 935 cr3 = read_cr3(); 936 937 for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) { 938 if (insn_table[cur_insn].supported_fn == NULL) 939 continue; 940 insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn(); 941 } 942 return VMX_TEST_START; 943 } 944 945 static void insn_intercept_main(void) 946 { 947 for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) { 948 vmx_set_test_stage(cur_insn * 2); 949 if ((insn_table[cur_insn].type == INSN_CPU0 && 950 !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) || 951 (insn_table[cur_insn].type == INSN_CPU1 && 952 !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) { 953 printf("\tCPU_CTRL%d.CPU_%s is not supported.\n", 954 insn_table[cur_insn].type - INSN_CPU0, 955 insn_table[cur_insn].name); 956 continue; 957 } 958 959 if (insn_table[cur_insn].disabled) { 960 printf("\tFeature required for %s is not supported.\n", 961 insn_table[cur_insn].name); 962 continue; 963 } 964 965 if ((insn_table[cur_insn].type == INSN_CPU0 && 966 !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) || 967 (insn_table[cur_insn].type == INSN_CPU1 && 968 !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) { 969 /* skip hlt, it stalls the guest and is tested below */ 970 if (insn_table[cur_insn].insn_func != insn_hlt) 971 insn_table[cur_insn].insn_func(); 972 report(vmx_get_test_stage() == cur_insn * 2, 973 "execute %s", 974 insn_table[cur_insn].name); 975 } else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP) 976 printf("\tCPU_CTRL%d.CPU_%s always traps.\n", 977 insn_table[cur_insn].type - INSN_CPU0, 978 insn_table[cur_insn].name); 979 980 vmcall(); 981 982 insn_table[cur_insn].insn_func(); 983 report(vmx_get_test_stage() == cur_insn * 2 + 1, 984 "intercept %s", 985 insn_table[cur_insn].name); 986 987 vmx_set_test_stage(cur_insn * 2 + 1); 988 vmcall(); 989 } 990 } 991 992 static int insn_intercept_exit_handler(void) 993 { 994 u64 guest_rip; 995 u32 reason; 996 ulong exit_qual; 997 u32 insn_len; 998 u32 insn_info; 999 bool pass; 1000 1001 guest_rip = vmcs_read(GUEST_RIP); 1002 reason = vmcs_read(EXI_REASON) & 0xff; 1003 exit_qual = vmcs_read(EXI_QUALIFICATION); 1004 insn_len = vmcs_read(EXI_INST_LEN); 1005 insn_info = vmcs_read(EXI_INST_INFO); 1006 1007 if (reason == VMX_VMCALL) { 1008 u32 val = 0; 1009 1010 if (insn_table[cur_insn].type == INSN_CPU0) 1011 val = vmcs_read(CPU_EXEC_CTRL0); 1012 else if (insn_table[cur_insn].type == INSN_CPU1) 1013 val = vmcs_read(CPU_EXEC_CTRL1); 1014 1015 if (vmx_get_test_stage() & 1) 1016 val &= ~insn_table[cur_insn].flag; 1017 else 1018 val |= insn_table[cur_insn].flag; 1019 1020 if (insn_table[cur_insn].type == INSN_CPU0) 1021 vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set); 1022 else if (insn_table[cur_insn].type == INSN_CPU1) 1023 vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set); 1024 } else { 1025 pass = (cur_insn * 2 == vmx_get_test_stage()) && 1026 insn_table[cur_insn].reason == reason; 1027 if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL && 1028 insn_table[cur_insn].exit_qual != exit_qual) 1029 pass = false; 1030 if (insn_table[cur_insn].test_field & FIELD_INSN_INFO && 1031 insn_table[cur_insn].insn_info != insn_info) 1032 pass = false; 1033 if (pass) 1034 vmx_inc_test_stage(); 1035 } 1036 vmcs_write(GUEST_RIP, guest_rip + insn_len); 1037 return VMX_TEST_RESUME; 1038 } 1039 1040 /** 1041 * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT) 1042 * @hpa: Host physical address of the top-level, a.k.a. root, EPT table 1043 * @enable_ad: Whether or not to enable Access/Dirty bits for EPT entries 1044 * 1045 * Returns 0 on success, 1 on failure. 1046 * 1047 * Note that @hpa doesn't need to point at actual memory if VM-Launch is 1048 * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy 1049 * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'. 1050 */ 1051 static int __setup_ept(u64 hpa, bool enable_ad) 1052 { 1053 if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) || 1054 !(ctrl_cpu_rev[1].clr & CPU_EPT)) { 1055 printf("\tEPT is not supported"); 1056 return 1; 1057 } 1058 if (!(ept_vpid.val & EPT_CAP_WB)) { 1059 printf("WB memtype for EPT walks not supported\n"); 1060 return 1; 1061 } 1062 if (!(ept_vpid.val & EPT_CAP_PWL4)) { 1063 printf("\tPWL4 is not supported\n"); 1064 return 1; 1065 } 1066 1067 eptp = EPT_MEM_TYPE_WB; 1068 eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT); 1069 eptp |= hpa; 1070 if (enable_ad) 1071 eptp |= EPTP_AD_FLAG; 1072 1073 vmcs_write(EPTP, eptp); 1074 vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY); 1075 vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT); 1076 1077 return 0; 1078 } 1079 1080 /** 1081 * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map 1082 * @enable_ad: Whether or not to enable Access/Dirty bits for EPT entries 1083 * 1084 * Returns 0 on success, 1 on failure. 1085 * 1086 * This is the "real" function for setting up EPT tables, i.e. use this for 1087 * tests that need to run code in the guest with EPT enabled. 1088 */ 1089 static int setup_ept(bool enable_ad) 1090 { 1091 unsigned long end_of_memory; 1092 1093 pml4 = alloc_page(); 1094 1095 if (__setup_ept(virt_to_phys(pml4), enable_ad)) 1096 return 1; 1097 1098 end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE); 1099 if (end_of_memory < (1ul << 32)) 1100 end_of_memory = (1ul << 32); 1101 /* Cannot use large EPT pages if we need to track EPT 1102 * accessed/dirty bits at 4K granularity. 1103 */ 1104 setup_ept_range(pml4, 0, end_of_memory, 0, 1105 !enable_ad && ept_2m_supported(), 1106 EPT_WA | EPT_RA | EPT_EA); 1107 return 0; 1108 } 1109 1110 /** 1111 * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA 1112 * 1113 * Setup EPT using a semi-arbitrary dummy root HPA. This function is intended 1114 * for use by tests that need EPT enabled to verify dependent VMCS controls 1115 * but never expect to fully enter the guest, i.e. don't need setup the actual 1116 * EPT tables. 1117 */ 1118 static void setup_dummy_ept(void) 1119 { 1120 if (__setup_ept(0, false)) 1121 report_abort("EPT setup unexpectedly failed"); 1122 } 1123 1124 static int enable_unrestricted_guest(void) 1125 { 1126 if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) || 1127 !(ctrl_cpu_rev[1].clr & CPU_URG) || 1128 !(ctrl_cpu_rev[1].clr & CPU_EPT)) 1129 return 1; 1130 1131 setup_dummy_ept(); 1132 1133 vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY); 1134 vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG); 1135 1136 return 0; 1137 } 1138 1139 static void ept_enable_ad_bits(void) 1140 { 1141 eptp |= EPTP_AD_FLAG; 1142 vmcs_write(EPTP, eptp); 1143 } 1144 1145 static void ept_disable_ad_bits(void) 1146 { 1147 eptp &= ~EPTP_AD_FLAG; 1148 vmcs_write(EPTP, eptp); 1149 } 1150 1151 static void ept_enable_ad_bits_or_skip_test(void) 1152 { 1153 if (!ept_ad_bits_supported()) 1154 test_skip("EPT AD bits not supported."); 1155 ept_enable_ad_bits(); 1156 } 1157 1158 static int apic_version; 1159 1160 static int ept_init_common(bool have_ad) 1161 { 1162 int ret; 1163 struct pci_dev pcidev; 1164 1165 if (setup_ept(have_ad)) 1166 return VMX_TEST_EXIT; 1167 data_page1 = alloc_page(); 1168 data_page2 = alloc_page(); 1169 *((u32 *)data_page1) = MAGIC_VAL_1; 1170 *((u32 *)data_page2) = MAGIC_VAL_2; 1171 install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2, 1172 EPT_RA | EPT_WA | EPT_EA); 1173 1174 apic_version = apic_read(APIC_LVR); 1175 1176 ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST); 1177 if (ret != PCIDEVADDR_INVALID) { 1178 pci_dev_init(&pcidev, ret); 1179 pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM]; 1180 } 1181 1182 return VMX_TEST_START; 1183 } 1184 1185 static int ept_init(struct vmcs *vmcs) 1186 { 1187 return ept_init_common(false); 1188 } 1189 1190 static void ept_common(void) 1191 { 1192 vmx_set_test_stage(0); 1193 if (*((u32 *)data_page2) != MAGIC_VAL_1 || 1194 *((u32 *)data_page1) != MAGIC_VAL_1) 1195 report(0, "EPT basic framework - read"); 1196 else { 1197 *((u32 *)data_page2) = MAGIC_VAL_3; 1198 vmcall(); 1199 if (vmx_get_test_stage() == 1) { 1200 if (*((u32 *)data_page1) == MAGIC_VAL_3 && 1201 *((u32 *)data_page2) == MAGIC_VAL_2) 1202 report(1, "EPT basic framework"); 1203 else 1204 report(1, "EPT basic framework - remap"); 1205 } 1206 } 1207 // Test EPT Misconfigurations 1208 vmx_set_test_stage(1); 1209 vmcall(); 1210 *((u32 *)data_page1) = MAGIC_VAL_1; 1211 if (vmx_get_test_stage() != 2) { 1212 report(0, "EPT misconfigurations"); 1213 goto t1; 1214 } 1215 vmx_set_test_stage(2); 1216 vmcall(); 1217 *((u32 *)data_page1) = MAGIC_VAL_1; 1218 report(vmx_get_test_stage() == 3, "EPT misconfigurations"); 1219 t1: 1220 // Test EPT violation 1221 vmx_set_test_stage(3); 1222 vmcall(); 1223 *((u32 *)data_page1) = MAGIC_VAL_1; 1224 report(vmx_get_test_stage() == 4, "EPT violation - page permission"); 1225 // Violation caused by EPT paging structure 1226 vmx_set_test_stage(4); 1227 vmcall(); 1228 *((u32 *)data_page1) = MAGIC_VAL_2; 1229 report(vmx_get_test_stage() == 5, "EPT violation - paging structure"); 1230 1231 // MMIO Read/Write 1232 vmx_set_test_stage(5); 1233 vmcall(); 1234 1235 *(u32 volatile *)pci_physaddr; 1236 report(vmx_get_test_stage() == 6, "MMIO EPT violation - read"); 1237 1238 *(u32 volatile *)pci_physaddr = MAGIC_VAL_1; 1239 report(vmx_get_test_stage() == 7, "MMIO EPT violation - write"); 1240 } 1241 1242 static void ept_main(void) 1243 { 1244 ept_common(); 1245 1246 // Test EPT access to L1 MMIO 1247 vmx_set_test_stage(7); 1248 report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access"); 1249 1250 // Test invalid operand for INVEPT 1251 vmcall(); 1252 report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT"); 1253 } 1254 1255 static bool invept_test(int type, u64 eptp) 1256 { 1257 bool ret, supported; 1258 1259 supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type); 1260 ret = invept(type, eptp); 1261 1262 if (ret == !supported) 1263 return false; 1264 1265 if (!supported) 1266 printf("WARNING: unsupported invept passed!\n"); 1267 else 1268 printf("WARNING: invept failed!\n"); 1269 1270 return true; 1271 } 1272 1273 static int pml_exit_handler(void) 1274 { 1275 u16 index, count; 1276 ulong reason = vmcs_read(EXI_REASON) & 0xff; 1277 u64 *pmlbuf = pml_log; 1278 u64 guest_rip = vmcs_read(GUEST_RIP);; 1279 u64 guest_cr3 = vmcs_read(GUEST_CR3); 1280 u32 insn_len = vmcs_read(EXI_INST_LEN); 1281 1282 switch (reason) { 1283 case VMX_VMCALL: 1284 switch (vmx_get_test_stage()) { 1285 case 0: 1286 index = vmcs_read(GUEST_PML_INDEX); 1287 for (count = index + 1; count < PML_INDEX; count++) { 1288 if (pmlbuf[count] == (u64)data_page2) { 1289 vmx_inc_test_stage(); 1290 clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2); 1291 break; 1292 } 1293 } 1294 break; 1295 case 1: 1296 index = vmcs_read(GUEST_PML_INDEX); 1297 /* Keep clearing the dirty bit till a overflow */ 1298 clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2); 1299 break; 1300 default: 1301 report(false, "unexpected stage, %d.", 1302 vmx_get_test_stage()); 1303 print_vmexit_info(); 1304 return VMX_TEST_VMEXIT; 1305 } 1306 vmcs_write(GUEST_RIP, guest_rip + insn_len); 1307 return VMX_TEST_RESUME; 1308 case VMX_PML_FULL: 1309 vmx_inc_test_stage(); 1310 vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1); 1311 return VMX_TEST_RESUME; 1312 default: 1313 report(false, "Unknown exit reason, %ld", reason); 1314 print_vmexit_info(); 1315 } 1316 return VMX_TEST_VMEXIT; 1317 } 1318 1319 static int ept_exit_handler_common(bool have_ad) 1320 { 1321 u64 guest_rip; 1322 u64 guest_cr3; 1323 ulong reason; 1324 u32 insn_len; 1325 u32 exit_qual; 1326 static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte; 1327 1328 guest_rip = vmcs_read(GUEST_RIP); 1329 guest_cr3 = vmcs_read(GUEST_CR3); 1330 reason = vmcs_read(EXI_REASON) & 0xff; 1331 insn_len = vmcs_read(EXI_INST_LEN); 1332 exit_qual = vmcs_read(EXI_QUALIFICATION); 1333 switch (reason) { 1334 case VMX_VMCALL: 1335 switch (vmx_get_test_stage()) { 1336 case 0: 1337 check_ept_ad(pml4, guest_cr3, 1338 (unsigned long)data_page1, 1339 have_ad ? EPT_ACCESS_FLAG : 0, 1340 have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0); 1341 check_ept_ad(pml4, guest_cr3, 1342 (unsigned long)data_page2, 1343 have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0, 1344 have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0); 1345 clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1); 1346 clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2); 1347 if (have_ad) 1348 ept_sync(INVEPT_SINGLE, eptp);; 1349 if (*((u32 *)data_page1) == MAGIC_VAL_3 && 1350 *((u32 *)data_page2) == MAGIC_VAL_2) { 1351 vmx_inc_test_stage(); 1352 install_ept(pml4, (unsigned long)data_page2, 1353 (unsigned long)data_page2, 1354 EPT_RA | EPT_WA | EPT_EA); 1355 } else 1356 report(0, "EPT basic framework - write"); 1357 break; 1358 case 1: 1359 install_ept(pml4, (unsigned long)data_page1, 1360 (unsigned long)data_page1, EPT_WA); 1361 ept_sync(INVEPT_SINGLE, eptp); 1362 break; 1363 case 2: 1364 install_ept(pml4, (unsigned long)data_page1, 1365 (unsigned long)data_page1, 1366 EPT_RA | EPT_WA | EPT_EA | 1367 (2 << EPT_MEM_TYPE_SHIFT)); 1368 ept_sync(INVEPT_SINGLE, eptp); 1369 break; 1370 case 3: 1371 clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1); 1372 TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1, 1373 1, &data_page1_pte)); 1374 set_ept_pte(pml4, (unsigned long)data_page1, 1375 1, data_page1_pte & ~EPT_PRESENT); 1376 ept_sync(INVEPT_SINGLE, eptp); 1377 break; 1378 case 4: 1379 TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1, 1380 2, &data_page1_pte)); 1381 data_page1_pte &= PAGE_MASK; 1382 TEST_ASSERT(get_ept_pte(pml4, data_page1_pte, 1383 2, &data_page1_pte_pte)); 1384 set_ept_pte(pml4, data_page1_pte, 2, 1385 data_page1_pte_pte & ~EPT_PRESENT); 1386 ept_sync(INVEPT_SINGLE, eptp); 1387 break; 1388 case 5: 1389 install_ept(pml4, (unsigned long)pci_physaddr, 1390 (unsigned long)pci_physaddr, 0); 1391 ept_sync(INVEPT_SINGLE, eptp); 1392 break; 1393 case 7: 1394 if (!invept_test(0, eptp)) 1395 vmx_inc_test_stage(); 1396 break; 1397 // Should not reach here 1398 default: 1399 report(false, "ERROR - unexpected stage, %d.", 1400 vmx_get_test_stage()); 1401 print_vmexit_info(); 1402 return VMX_TEST_VMEXIT; 1403 } 1404 vmcs_write(GUEST_RIP, guest_rip + insn_len); 1405 return VMX_TEST_RESUME; 1406 case VMX_EPT_MISCONFIG: 1407 switch (vmx_get_test_stage()) { 1408 case 1: 1409 case 2: 1410 vmx_inc_test_stage(); 1411 install_ept(pml4, (unsigned long)data_page1, 1412 (unsigned long)data_page1, 1413 EPT_RA | EPT_WA | EPT_EA); 1414 ept_sync(INVEPT_SINGLE, eptp); 1415 break; 1416 // Should not reach here 1417 default: 1418 report(false, "ERROR - unexpected stage, %d.", 1419 vmx_get_test_stage()); 1420 print_vmexit_info(); 1421 return VMX_TEST_VMEXIT; 1422 } 1423 return VMX_TEST_RESUME; 1424 case VMX_EPT_VIOLATION: 1425 /* 1426 * Exit-qualifications are masked not to account for advanced 1427 * VM-exit information. Once KVM supports this feature, this 1428 * masking should be removed. 1429 */ 1430 exit_qual &= ~EPT_VLT_GUEST_MASK; 1431 1432 switch(vmx_get_test_stage()) { 1433 case 3: 1434 check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0, 1435 have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0); 1436 clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1); 1437 if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD | 1438 EPT_VLT_PADDR)) 1439 vmx_inc_test_stage(); 1440 set_ept_pte(pml4, (unsigned long)data_page1, 1441 1, data_page1_pte | (EPT_PRESENT)); 1442 ept_sync(INVEPT_SINGLE, eptp); 1443 break; 1444 case 4: 1445 check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0, 1446 have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0); 1447 clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1); 1448 if (exit_qual == (EPT_VLT_RD | 1449 (have_ad ? EPT_VLT_WR : 0) | 1450 EPT_VLT_LADDR_VLD)) 1451 vmx_inc_test_stage(); 1452 set_ept_pte(pml4, data_page1_pte, 2, 1453 data_page1_pte_pte | (EPT_PRESENT)); 1454 ept_sync(INVEPT_SINGLE, eptp); 1455 break; 1456 case 5: 1457 if (exit_qual & EPT_VLT_RD) 1458 vmx_inc_test_stage(); 1459 TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr, 1460 1, &memaddr_pte)); 1461 set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA); 1462 ept_sync(INVEPT_SINGLE, eptp); 1463 break; 1464 case 6: 1465 if (exit_qual & EPT_VLT_WR) 1466 vmx_inc_test_stage(); 1467 TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr, 1468 1, &memaddr_pte)); 1469 set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA); 1470 ept_sync(INVEPT_SINGLE, eptp); 1471 break; 1472 default: 1473 // Should not reach here 1474 report(false, "ERROR : unexpected stage, %d", 1475 vmx_get_test_stage()); 1476 print_vmexit_info(); 1477 return VMX_TEST_VMEXIT; 1478 } 1479 return VMX_TEST_RESUME; 1480 default: 1481 report(false, "Unknown exit reason, %ld", reason); 1482 print_vmexit_info(); 1483 } 1484 return VMX_TEST_VMEXIT; 1485 } 1486 1487 static int ept_exit_handler(void) 1488 { 1489 return ept_exit_handler_common(false); 1490 } 1491 1492 static int eptad_init(struct vmcs *vmcs) 1493 { 1494 int r = ept_init_common(true); 1495 1496 if (r == VMX_TEST_EXIT) 1497 return r; 1498 1499 if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) { 1500 printf("\tEPT A/D bits are not supported"); 1501 return VMX_TEST_EXIT; 1502 } 1503 1504 return r; 1505 } 1506 1507 static int pml_init(struct vmcs *vmcs) 1508 { 1509 u32 ctrl_cpu; 1510 int r = eptad_init(vmcs); 1511 1512 if (r == VMX_TEST_EXIT) 1513 return r; 1514 1515 if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) || 1516 !(ctrl_cpu_rev[1].clr & CPU_PML)) { 1517 printf("\tPML is not supported"); 1518 return VMX_TEST_EXIT; 1519 } 1520 1521 pml_log = alloc_page(); 1522 vmcs_write(PMLADDR, (u64)pml_log); 1523 vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1); 1524 1525 ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML; 1526 vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu); 1527 1528 return VMX_TEST_START; 1529 } 1530 1531 static void pml_main(void) 1532 { 1533 int count = 0; 1534 1535 vmx_set_test_stage(0); 1536 *((u32 *)data_page2) = 0x1; 1537 vmcall(); 1538 report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging"); 1539 1540 while (vmx_get_test_stage() == 1) { 1541 vmcall(); 1542 *((u32 *)data_page2) = 0x1; 1543 if (count++ > PML_INDEX) 1544 break; 1545 } 1546 report(vmx_get_test_stage() == 2, "PML Full Event"); 1547 } 1548 1549 static void eptad_main(void) 1550 { 1551 ept_common(); 1552 } 1553 1554 static int eptad_exit_handler(void) 1555 { 1556 return ept_exit_handler_common(true); 1557 } 1558 1559 static bool invvpid_test(int type, u16 vpid) 1560 { 1561 bool ret, supported; 1562 1563 supported = ept_vpid.val & 1564 (VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type); 1565 ret = invvpid(type, vpid, 0); 1566 1567 if (ret == !supported) 1568 return false; 1569 1570 if (!supported) 1571 printf("WARNING: unsupported invvpid passed!\n"); 1572 else 1573 printf("WARNING: invvpid failed!\n"); 1574 1575 return true; 1576 } 1577 1578 static int vpid_init(struct vmcs *vmcs) 1579 { 1580 u32 ctrl_cpu1; 1581 1582 if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) || 1583 !(ctrl_cpu_rev[1].clr & CPU_VPID)) { 1584 printf("\tVPID is not supported"); 1585 return VMX_TEST_EXIT; 1586 } 1587 1588 ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1); 1589 ctrl_cpu1 |= CPU_VPID; 1590 vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1); 1591 return VMX_TEST_START; 1592 } 1593 1594 static void vpid_main(void) 1595 { 1596 vmx_set_test_stage(0); 1597 vmcall(); 1598 report(vmx_get_test_stage() == 1, "INVVPID SINGLE ADDRESS"); 1599 vmx_set_test_stage(2); 1600 vmcall(); 1601 report(vmx_get_test_stage() == 3, "INVVPID SINGLE"); 1602 vmx_set_test_stage(4); 1603 vmcall(); 1604 report(vmx_get_test_stage() == 5, "INVVPID ALL"); 1605 } 1606 1607 static int vpid_exit_handler(void) 1608 { 1609 u64 guest_rip; 1610 ulong reason; 1611 u32 insn_len; 1612 1613 guest_rip = vmcs_read(GUEST_RIP); 1614 reason = vmcs_read(EXI_REASON) & 0xff; 1615 insn_len = vmcs_read(EXI_INST_LEN); 1616 1617 switch (reason) { 1618 case VMX_VMCALL: 1619 switch(vmx_get_test_stage()) { 1620 case 0: 1621 if (!invvpid_test(INVVPID_ADDR, 1)) 1622 vmx_inc_test_stage(); 1623 break; 1624 case 2: 1625 if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1)) 1626 vmx_inc_test_stage(); 1627 break; 1628 case 4: 1629 if (!invvpid_test(INVVPID_ALL, 1)) 1630 vmx_inc_test_stage(); 1631 break; 1632 default: 1633 report(false, "ERROR: unexpected stage, %d", 1634 vmx_get_test_stage()); 1635 print_vmexit_info(); 1636 return VMX_TEST_VMEXIT; 1637 } 1638 vmcs_write(GUEST_RIP, guest_rip + insn_len); 1639 return VMX_TEST_RESUME; 1640 default: 1641 report(false, "Unknown exit reason, %ld", reason); 1642 print_vmexit_info(); 1643 } 1644 return VMX_TEST_VMEXIT; 1645 } 1646 1647 #define TIMER_VECTOR 222 1648 1649 static volatile bool timer_fired; 1650 1651 static void timer_isr(isr_regs_t *regs) 1652 { 1653 timer_fired = true; 1654 apic_write(APIC_EOI, 0); 1655 } 1656 1657 static int interrupt_init(struct vmcs *vmcs) 1658 { 1659 msr_bmp_init(); 1660 vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT); 1661 handle_irq(TIMER_VECTOR, timer_isr); 1662 return VMX_TEST_START; 1663 } 1664 1665 static void interrupt_main(void) 1666 { 1667 long long start, loops; 1668 1669 vmx_set_test_stage(0); 1670 1671 apic_write(APIC_LVTT, TIMER_VECTOR); 1672 irq_enable(); 1673 1674 apic_write(APIC_TMICT, 1); 1675 for (loops = 0; loops < 10000000 && !timer_fired; loops++) 1676 asm volatile ("nop"); 1677 report(timer_fired, "direct interrupt while running guest"); 1678 1679 apic_write(APIC_TMICT, 0); 1680 irq_disable(); 1681 vmcall(); 1682 timer_fired = false; 1683 apic_write(APIC_TMICT, 1); 1684 for (loops = 0; loops < 10000000 && !timer_fired; loops++) 1685 asm volatile ("nop"); 1686 report(timer_fired, "intercepted interrupt while running guest"); 1687 1688 irq_enable(); 1689 apic_write(APIC_TMICT, 0); 1690 irq_disable(); 1691 vmcall(); 1692 timer_fired = false; 1693 start = rdtsc(); 1694 apic_write(APIC_TMICT, 1000000); 1695 1696 asm volatile ("sti; hlt"); 1697 1698 report(rdtsc() - start > 1000000 && timer_fired, 1699 "direct interrupt + hlt"); 1700 1701 apic_write(APIC_TMICT, 0); 1702 irq_disable(); 1703 vmcall(); 1704 timer_fired = false; 1705 start = rdtsc(); 1706 apic_write(APIC_TMICT, 1000000); 1707 1708 asm volatile ("sti; hlt"); 1709 1710 report(rdtsc() - start > 10000 && timer_fired, 1711 "intercepted interrupt + hlt"); 1712 1713 apic_write(APIC_TMICT, 0); 1714 irq_disable(); 1715 vmcall(); 1716 timer_fired = false; 1717 start = rdtsc(); 1718 apic_write(APIC_TMICT, 1000000); 1719 1720 irq_enable(); 1721 asm volatile ("nop"); 1722 vmcall(); 1723 1724 report(rdtsc() - start > 10000 && timer_fired, 1725 "direct interrupt + activity state hlt"); 1726 1727 apic_write(APIC_TMICT, 0); 1728 irq_disable(); 1729 vmcall(); 1730 timer_fired = false; 1731 start = rdtsc(); 1732 apic_write(APIC_TMICT, 1000000); 1733 1734 irq_enable(); 1735 asm volatile ("nop"); 1736 vmcall(); 1737 1738 report(rdtsc() - start > 10000 && timer_fired, 1739 "intercepted interrupt + activity state hlt"); 1740 1741 apic_write(APIC_TMICT, 0); 1742 irq_disable(); 1743 vmx_set_test_stage(7); 1744 vmcall(); 1745 timer_fired = false; 1746 apic_write(APIC_TMICT, 1); 1747 for (loops = 0; loops < 10000000 && !timer_fired; loops++) 1748 asm volatile ("nop"); 1749 report(timer_fired, 1750 "running a guest with interrupt acknowledgement set"); 1751 1752 apic_write(APIC_TMICT, 0); 1753 irq_enable(); 1754 timer_fired = false; 1755 vmcall(); 1756 report(timer_fired, "Inject an event to a halted guest"); 1757 } 1758 1759 static int interrupt_exit_handler(void) 1760 { 1761 u64 guest_rip = vmcs_read(GUEST_RIP); 1762 ulong reason = vmcs_read(EXI_REASON) & 0xff; 1763 u32 insn_len = vmcs_read(EXI_INST_LEN); 1764 1765 switch (reason) { 1766 case VMX_VMCALL: 1767 switch (vmx_get_test_stage()) { 1768 case 0: 1769 case 2: 1770 case 5: 1771 vmcs_write(PIN_CONTROLS, 1772 vmcs_read(PIN_CONTROLS) | PIN_EXTINT); 1773 break; 1774 case 7: 1775 vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA); 1776 vmcs_write(PIN_CONTROLS, 1777 vmcs_read(PIN_CONTROLS) | PIN_EXTINT); 1778 break; 1779 case 1: 1780 case 3: 1781 vmcs_write(PIN_CONTROLS, 1782 vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT); 1783 break; 1784 case 4: 1785 case 6: 1786 vmcs_write(GUEST_ACTV_STATE, ACTV_HLT); 1787 break; 1788 1789 case 8: 1790 vmcs_write(GUEST_ACTV_STATE, ACTV_HLT); 1791 vmcs_write(ENT_INTR_INFO, 1792 TIMER_VECTOR | 1793 (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) | 1794 INTR_INFO_VALID_MASK); 1795 break; 1796 } 1797 vmx_inc_test_stage(); 1798 vmcs_write(GUEST_RIP, guest_rip + insn_len); 1799 return VMX_TEST_RESUME; 1800 case VMX_EXTINT: 1801 if (vmcs_read(EXI_CONTROLS) & EXI_INTA) { 1802 int vector = vmcs_read(EXI_INTR_INFO) & 0xff; 1803 handle_external_interrupt(vector); 1804 } else { 1805 irq_enable(); 1806 asm volatile ("nop"); 1807 irq_disable(); 1808 } 1809 if (vmx_get_test_stage() >= 2) 1810 vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE); 1811 return VMX_TEST_RESUME; 1812 default: 1813 report(false, "Unknown exit reason, %ld", reason); 1814 print_vmexit_info(); 1815 } 1816 1817 return VMX_TEST_VMEXIT; 1818 } 1819 1820 static int dbgctls_init(struct vmcs *vmcs) 1821 { 1822 u64 dr7 = 0x402; 1823 u64 zero = 0; 1824 1825 msr_bmp_init(); 1826 asm volatile( 1827 "mov %0,%%dr0\n\t" 1828 "mov %0,%%dr1\n\t" 1829 "mov %0,%%dr2\n\t" 1830 "mov %1,%%dr7\n\t" 1831 : : "r" (zero), "r" (dr7)); 1832 wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1); 1833 vmcs_write(GUEST_DR7, 0x404); 1834 vmcs_write(GUEST_DEBUGCTL, 0x2); 1835 1836 vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS); 1837 vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS); 1838 1839 return VMX_TEST_START; 1840 } 1841 1842 static void dbgctls_main(void) 1843 { 1844 u64 dr7, debugctl; 1845 1846 asm volatile("mov %%dr7,%0" : "=r" (dr7)); 1847 debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR); 1848 /* Commented out: KVM does not support DEBUGCTL so far */ 1849 (void)debugctl; 1850 report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */); 1851 1852 dr7 = 0x408; 1853 asm volatile("mov %0,%%dr7" : : "r" (dr7)); 1854 wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3); 1855 1856 vmx_set_test_stage(0); 1857 vmcall(); 1858 report(vmx_get_test_stage() == 1, "Save debug controls"); 1859 1860 if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS || 1861 ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) { 1862 printf("\tDebug controls are always loaded/saved\n"); 1863 return; 1864 } 1865 vmx_set_test_stage(2); 1866 vmcall(); 1867 1868 asm volatile("mov %%dr7,%0" : "=r" (dr7)); 1869 debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR); 1870 /* Commented out: KVM does not support DEBUGCTL so far */ 1871 (void)debugctl; 1872 report(dr7 == 0x402, 1873 "Guest=host debug controls" /* && debugctl == 0x1 */); 1874 1875 dr7 = 0x408; 1876 asm volatile("mov %0,%%dr7" : : "r" (dr7)); 1877 wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3); 1878 1879 vmx_set_test_stage(3); 1880 vmcall(); 1881 report(vmx_get_test_stage() == 4, "Don't save debug controls"); 1882 } 1883 1884 static int dbgctls_exit_handler(void) 1885 { 1886 unsigned int reason = vmcs_read(EXI_REASON) & 0xff; 1887 u32 insn_len = vmcs_read(EXI_INST_LEN); 1888 u64 guest_rip = vmcs_read(GUEST_RIP); 1889 u64 dr7, debugctl; 1890 1891 asm volatile("mov %%dr7,%0" : "=r" (dr7)); 1892 debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR); 1893 1894 switch (reason) { 1895 case VMX_VMCALL: 1896 switch (vmx_get_test_stage()) { 1897 case 0: 1898 if (dr7 == 0x400 && debugctl == 0 && 1899 vmcs_read(GUEST_DR7) == 0x408 /* && 1900 Commented out: KVM does not support DEBUGCTL so far 1901 vmcs_read(GUEST_DEBUGCTL) == 0x3 */) 1902 vmx_inc_test_stage(); 1903 break; 1904 case 2: 1905 dr7 = 0x402; 1906 asm volatile("mov %0,%%dr7" : : "r" (dr7)); 1907 wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1); 1908 vmcs_write(GUEST_DR7, 0x404); 1909 vmcs_write(GUEST_DEBUGCTL, 0x2); 1910 1911 vmcs_write(ENT_CONTROLS, 1912 vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS); 1913 vmcs_write(EXI_CONTROLS, 1914 vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS); 1915 break; 1916 case 3: 1917 if (dr7 == 0x400 && debugctl == 0 && 1918 vmcs_read(GUEST_DR7) == 0x404 /* && 1919 Commented out: KVM does not support DEBUGCTL so far 1920 vmcs_read(GUEST_DEBUGCTL) == 0x2 */) 1921 vmx_inc_test_stage(); 1922 break; 1923 } 1924 vmcs_write(GUEST_RIP, guest_rip + insn_len); 1925 return VMX_TEST_RESUME; 1926 default: 1927 report(false, "Unknown exit reason, %d", reason); 1928 print_vmexit_info(); 1929 } 1930 return VMX_TEST_VMEXIT; 1931 } 1932 1933 struct vmx_msr_entry { 1934 u32 index; 1935 u32 reserved; 1936 u64 value; 1937 } __attribute__((packed)); 1938 1939 #define MSR_MAGIC 0x31415926 1940 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load; 1941 1942 static int msr_switch_init(struct vmcs *vmcs) 1943 { 1944 msr_bmp_init(); 1945 exit_msr_store = alloc_page(); 1946 exit_msr_load = alloc_page(); 1947 entry_msr_load = alloc_page(); 1948 entry_msr_load[0].index = MSR_KERNEL_GS_BASE; 1949 entry_msr_load[0].value = MSR_MAGIC; 1950 1951 vmx_set_test_stage(1); 1952 vmcs_write(ENT_MSR_LD_CNT, 1); 1953 vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load); 1954 vmcs_write(EXI_MSR_ST_CNT, 1); 1955 vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store); 1956 vmcs_write(EXI_MSR_LD_CNT, 1); 1957 vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load); 1958 return VMX_TEST_START; 1959 } 1960 1961 static void msr_switch_main(void) 1962 { 1963 if (vmx_get_test_stage() == 1) { 1964 report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC, 1965 "VM entry MSR load"); 1966 vmx_set_test_stage(2); 1967 wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1); 1968 exit_msr_store[0].index = MSR_KERNEL_GS_BASE; 1969 exit_msr_load[0].index = MSR_KERNEL_GS_BASE; 1970 exit_msr_load[0].value = MSR_MAGIC + 2; 1971 } 1972 vmcall(); 1973 } 1974 1975 static int msr_switch_exit_handler(void) 1976 { 1977 ulong reason; 1978 1979 reason = vmcs_read(EXI_REASON); 1980 if (reason == VMX_VMCALL && vmx_get_test_stage() == 2) { 1981 report(exit_msr_store[0].value == MSR_MAGIC + 1, 1982 "VM exit MSR store"); 1983 report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2, 1984 "VM exit MSR load"); 1985 vmx_set_test_stage(3); 1986 entry_msr_load[0].index = MSR_FS_BASE; 1987 return VMX_TEST_RESUME; 1988 } 1989 printf("ERROR %s: unexpected stage=%u or reason=%lu\n", 1990 __func__, vmx_get_test_stage(), reason); 1991 return VMX_TEST_EXIT; 1992 } 1993 1994 static int msr_switch_entry_failure(struct vmentry_failure *failure) 1995 { 1996 ulong reason; 1997 1998 if (failure->early) { 1999 printf("ERROR %s: early exit\n", __func__); 2000 return VMX_TEST_EXIT; 2001 } 2002 2003 reason = vmcs_read(EXI_REASON); 2004 if (reason == (VMX_ENTRY_FAILURE | VMX_FAIL_MSR) && 2005 vmx_get_test_stage() == 3) { 2006 report(vmcs_read(EXI_QUALIFICATION) == 1, 2007 "VM entry MSR load: try to load FS_BASE"); 2008 return VMX_TEST_VMEXIT; 2009 } 2010 printf("ERROR %s: unexpected stage=%u or reason=%lu\n", 2011 __func__, vmx_get_test_stage(), reason); 2012 return VMX_TEST_EXIT; 2013 } 2014 2015 static int vmmcall_init(struct vmcs *vmcs) 2016 { 2017 vmcs_write(EXC_BITMAP, 1 << UD_VECTOR); 2018 return VMX_TEST_START; 2019 } 2020 2021 static void vmmcall_main(void) 2022 { 2023 asm volatile( 2024 "mov $0xABCD, %%rax\n\t" 2025 "vmmcall\n\t" 2026 ::: "rax"); 2027 2028 report(0, "VMMCALL"); 2029 } 2030 2031 static int vmmcall_exit_handler(void) 2032 { 2033 ulong reason; 2034 2035 reason = vmcs_read(EXI_REASON); 2036 switch (reason) { 2037 case VMX_VMCALL: 2038 printf("here\n"); 2039 report(0, "VMMCALL triggers #UD"); 2040 break; 2041 case VMX_EXC_NMI: 2042 report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR, 2043 "VMMCALL triggers #UD"); 2044 break; 2045 default: 2046 report(false, "Unknown exit reason, %ld", reason); 2047 print_vmexit_info(); 2048 } 2049 2050 return VMX_TEST_VMEXIT; 2051 } 2052 2053 static int disable_rdtscp_init(struct vmcs *vmcs) 2054 { 2055 u32 ctrl_cpu1; 2056 2057 if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) { 2058 ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1); 2059 ctrl_cpu1 &= ~CPU_RDTSCP; 2060 vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1); 2061 } 2062 2063 return VMX_TEST_START; 2064 } 2065 2066 static void disable_rdtscp_ud_handler(struct ex_regs *regs) 2067 { 2068 switch (vmx_get_test_stage()) { 2069 case 0: 2070 report(true, "RDTSCP triggers #UD"); 2071 vmx_inc_test_stage(); 2072 regs->rip += 3; 2073 break; 2074 case 2: 2075 report(true, "RDPID triggers #UD"); 2076 vmx_inc_test_stage(); 2077 regs->rip += 4; 2078 break; 2079 } 2080 return; 2081 2082 } 2083 2084 static void disable_rdtscp_main(void) 2085 { 2086 /* Test that #UD is properly injected in L2. */ 2087 handle_exception(UD_VECTOR, disable_rdtscp_ud_handler); 2088 2089 vmx_set_test_stage(0); 2090 asm volatile("rdtscp" : : : "eax", "ecx", "edx"); 2091 vmcall(); 2092 asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax"); 2093 2094 handle_exception(UD_VECTOR, 0); 2095 vmcall(); 2096 } 2097 2098 static int disable_rdtscp_exit_handler(void) 2099 { 2100 unsigned int reason = vmcs_read(EXI_REASON) & 0xff; 2101 2102 switch (reason) { 2103 case VMX_VMCALL: 2104 switch (vmx_get_test_stage()) { 2105 case 0: 2106 report(false, "RDTSCP triggers #UD"); 2107 vmx_inc_test_stage(); 2108 /* fallthrough */ 2109 case 1: 2110 vmx_inc_test_stage(); 2111 vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3); 2112 return VMX_TEST_RESUME; 2113 case 2: 2114 report(false, "RDPID triggers #UD"); 2115 break; 2116 } 2117 break; 2118 2119 default: 2120 report(false, "Unknown exit reason, %d", reason); 2121 print_vmexit_info(); 2122 } 2123 return VMX_TEST_VMEXIT; 2124 } 2125 2126 static int int3_init(struct vmcs *vmcs) 2127 { 2128 vmcs_write(EXC_BITMAP, ~0u); 2129 return VMX_TEST_START; 2130 } 2131 2132 static void int3_guest_main(void) 2133 { 2134 asm volatile ("int3"); 2135 } 2136 2137 static int int3_exit_handler(void) 2138 { 2139 u32 reason = vmcs_read(EXI_REASON); 2140 u32 intr_info = vmcs_read(EXI_INTR_INFO); 2141 2142 report(reason == VMX_EXC_NMI && (intr_info & INTR_INFO_VALID_MASK) && 2143 (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR && 2144 ((intr_info & INTR_INFO_INTR_TYPE_MASK) >> 2145 INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION, 2146 "L1 intercepts #BP"); 2147 2148 return VMX_TEST_VMEXIT; 2149 } 2150 2151 static int into_init(struct vmcs *vmcs) 2152 { 2153 vmcs_write(EXC_BITMAP, ~0u); 2154 return VMX_TEST_START; 2155 } 2156 2157 static void into_guest_main(void) 2158 { 2159 struct far_pointer32 fp = { 2160 .offset = (uintptr_t)&&into, 2161 .selector = KERNEL_CS32, 2162 }; 2163 register uintptr_t rsp asm("rsp"); 2164 2165 if (fp.offset != (uintptr_t)&&into) { 2166 printf("Code address too high.\n"); 2167 return; 2168 } 2169 if ((u32)rsp != rsp) { 2170 printf("Stack address too high.\n"); 2171 return; 2172 } 2173 2174 asm goto ("lcall *%0" : : "m" (fp) : "rax" : into); 2175 return; 2176 into: 2177 asm volatile (".code32;" 2178 "movl $0x7fffffff, %eax;" 2179 "addl %eax, %eax;" 2180 "into;" 2181 "lret;" 2182 ".code64"); 2183 __builtin_unreachable(); 2184 } 2185 2186 static int into_exit_handler(void) 2187 { 2188 u32 reason = vmcs_read(EXI_REASON); 2189 u32 intr_info = vmcs_read(EXI_INTR_INFO); 2190 2191 report(reason == VMX_EXC_NMI && (intr_info & INTR_INFO_VALID_MASK) && 2192 (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR && 2193 ((intr_info & INTR_INFO_INTR_TYPE_MASK) >> 2194 INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION, 2195 "L1 intercepts #OF"); 2196 2197 return VMX_TEST_VMEXIT; 2198 } 2199 2200 static void exit_monitor_from_l2_main(void) 2201 { 2202 printf("Calling exit(0) from l2...\n"); 2203 exit(0); 2204 } 2205 2206 static int exit_monitor_from_l2_handler(void) 2207 { 2208 report(false, "The guest should have killed the VMM"); 2209 return VMX_TEST_EXIT; 2210 } 2211 2212 static void assert_exit_reason(u64 expected) 2213 { 2214 u64 actual = vmcs_read(EXI_REASON); 2215 2216 TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.", 2217 exit_reason_description(expected), 2218 exit_reason_description(actual)); 2219 } 2220 2221 static void skip_exit_insn(void) 2222 { 2223 u64 guest_rip = vmcs_read(GUEST_RIP); 2224 u32 insn_len = vmcs_read(EXI_INST_LEN); 2225 vmcs_write(GUEST_RIP, guest_rip + insn_len); 2226 } 2227 2228 static void skip_exit_vmcall(void) 2229 { 2230 assert_exit_reason(VMX_VMCALL); 2231 skip_exit_insn(); 2232 } 2233 2234 static void v2_null_test_guest(void) 2235 { 2236 } 2237 2238 static void v2_null_test(void) 2239 { 2240 test_set_guest(v2_null_test_guest); 2241 enter_guest(); 2242 report(1, __func__); 2243 } 2244 2245 static void v2_multiple_entries_test_guest(void) 2246 { 2247 vmx_set_test_stage(1); 2248 vmcall(); 2249 vmx_set_test_stage(2); 2250 } 2251 2252 static void v2_multiple_entries_test(void) 2253 { 2254 test_set_guest(v2_multiple_entries_test_guest); 2255 enter_guest(); 2256 TEST_ASSERT_EQ(vmx_get_test_stage(), 1); 2257 skip_exit_vmcall(); 2258 enter_guest(); 2259 TEST_ASSERT_EQ(vmx_get_test_stage(), 2); 2260 report(1, __func__); 2261 } 2262 2263 static int fixture_test_data = 1; 2264 2265 static void fixture_test_teardown(void *data) 2266 { 2267 *((int *) data) = 1; 2268 } 2269 2270 static void fixture_test_guest(void) 2271 { 2272 fixture_test_data++; 2273 } 2274 2275 2276 static void fixture_test_setup(void) 2277 { 2278 TEST_ASSERT_EQ_MSG(1, fixture_test_data, 2279 "fixture_test_teardown didn't run?!"); 2280 fixture_test_data = 2; 2281 test_add_teardown(fixture_test_teardown, &fixture_test_data); 2282 test_set_guest(fixture_test_guest); 2283 } 2284 2285 static void fixture_test_case1(void) 2286 { 2287 fixture_test_setup(); 2288 TEST_ASSERT_EQ(2, fixture_test_data); 2289 enter_guest(); 2290 TEST_ASSERT_EQ(3, fixture_test_data); 2291 report(1, __func__); 2292 } 2293 2294 static void fixture_test_case2(void) 2295 { 2296 fixture_test_setup(); 2297 TEST_ASSERT_EQ(2, fixture_test_data); 2298 enter_guest(); 2299 TEST_ASSERT_EQ(3, fixture_test_data); 2300 report(1, __func__); 2301 } 2302 2303 enum ept_access_op { 2304 OP_READ, 2305 OP_WRITE, 2306 OP_EXEC, 2307 OP_FLUSH_TLB, 2308 OP_EXIT, 2309 }; 2310 2311 static struct ept_access_test_data { 2312 unsigned long gpa; 2313 unsigned long *gva; 2314 unsigned long hpa; 2315 unsigned long *hva; 2316 enum ept_access_op op; 2317 } ept_access_test_data; 2318 2319 extern unsigned char ret42_start; 2320 extern unsigned char ret42_end; 2321 2322 /* Returns 42. */ 2323 asm( 2324 ".align 64\n" 2325 "ret42_start:\n" 2326 "mov $42, %eax\n" 2327 "ret\n" 2328 "ret42_end:\n" 2329 ); 2330 2331 static void 2332 diagnose_ept_violation_qual(u64 expected, u64 actual) 2333 { 2334 2335 #define DIAGNOSE(flag) \ 2336 do { \ 2337 if ((expected & flag) != (actual & flag)) \ 2338 printf(#flag " %sexpected\n", \ 2339 (expected & flag) ? "" : "un"); \ 2340 } while (0) 2341 2342 DIAGNOSE(EPT_VLT_RD); 2343 DIAGNOSE(EPT_VLT_WR); 2344 DIAGNOSE(EPT_VLT_FETCH); 2345 DIAGNOSE(EPT_VLT_PERM_RD); 2346 DIAGNOSE(EPT_VLT_PERM_WR); 2347 DIAGNOSE(EPT_VLT_PERM_EX); 2348 DIAGNOSE(EPT_VLT_LADDR_VLD); 2349 DIAGNOSE(EPT_VLT_PADDR); 2350 2351 #undef DIAGNOSE 2352 } 2353 2354 static void do_ept_access_op(enum ept_access_op op) 2355 { 2356 ept_access_test_data.op = op; 2357 enter_guest(); 2358 } 2359 2360 /* 2361 * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only 2362 * needed by tests that modify guest PTEs. 2363 */ 2364 static void ept_access_test_guest_flush_tlb(void) 2365 { 2366 do_ept_access_op(OP_FLUSH_TLB); 2367 skip_exit_vmcall(); 2368 } 2369 2370 /* 2371 * Modifies the EPT entry at @level in the mapping of @gpa. First clears the 2372 * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into 2373 * a huge page. 2374 */ 2375 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level, 2376 unsigned long clear, unsigned long set) 2377 { 2378 struct ept_access_test_data *data = &ept_access_test_data; 2379 unsigned long orig_pte; 2380 unsigned long pte; 2381 2382 /* Screw with the mapping at the requested level. */ 2383 TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte)); 2384 pte = orig_pte; 2385 if (mkhuge) 2386 pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE; 2387 else 2388 pte = orig_pte; 2389 pte = (pte & ~clear) | set; 2390 set_ept_pte(pml4, gpa, level, pte); 2391 ept_sync(INVEPT_SINGLE, eptp); 2392 2393 return orig_pte; 2394 } 2395 2396 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte) 2397 { 2398 set_ept_pte(pml4, gpa, level, orig_pte); 2399 ept_sync(INVEPT_SINGLE, eptp); 2400 } 2401 2402 static void do_ept_violation(bool leaf, enum ept_access_op op, 2403 u64 expected_qual, u64 expected_paddr) 2404 { 2405 u64 qual; 2406 2407 /* Try the access and observe the violation. */ 2408 do_ept_access_op(op); 2409 2410 assert_exit_reason(VMX_EPT_VIOLATION); 2411 2412 qual = vmcs_read(EXI_QUALIFICATION); 2413 2414 /* Mask undefined bits (which may later be defined in certain cases). */ 2415 qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX | 2416 EPT_VLT_PERM_USER_EX); 2417 2418 diagnose_ept_violation_qual(expected_qual, qual); 2419 TEST_EXPECT_EQ(expected_qual, qual); 2420 2421 #if 0 2422 /* Disable for now otherwise every test will fail */ 2423 TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS), 2424 (unsigned long) ( 2425 op == OP_EXEC ? data->gva + 1 : data->gva)); 2426 #endif 2427 /* 2428 * TODO: tests that probe expected_paddr in pages other than the one at 2429 * the beginning of the 1g region. 2430 */ 2431 TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr); 2432 } 2433 2434 static void 2435 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear, 2436 unsigned long set, enum ept_access_op op, 2437 u64 expected_qual) 2438 { 2439 struct ept_access_test_data *data = &ept_access_test_data; 2440 unsigned long orig_pte; 2441 2442 orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set); 2443 2444 do_ept_violation(level == 1 || mkhuge, op, expected_qual, 2445 op == OP_EXEC ? data->gpa + sizeof(unsigned long) : 2446 data->gpa); 2447 2448 /* Fix the violation and resume the op loop. */ 2449 ept_untwiddle(data->gpa, level, orig_pte); 2450 enter_guest(); 2451 skip_exit_vmcall(); 2452 } 2453 2454 static void 2455 ept_violation_at_level(int level, unsigned long clear, unsigned long set, 2456 enum ept_access_op op, u64 expected_qual) 2457 { 2458 ept_violation_at_level_mkhuge(false, level, clear, set, op, 2459 expected_qual); 2460 if (ept_huge_pages_supported(level)) 2461 ept_violation_at_level_mkhuge(true, level, clear, set, op, 2462 expected_qual); 2463 } 2464 2465 static void ept_violation(unsigned long clear, unsigned long set, 2466 enum ept_access_op op, u64 expected_qual) 2467 { 2468 ept_violation_at_level(1, clear, set, op, expected_qual); 2469 ept_violation_at_level(2, clear, set, op, expected_qual); 2470 ept_violation_at_level(3, clear, set, op, expected_qual); 2471 ept_violation_at_level(4, clear, set, op, expected_qual); 2472 } 2473 2474 static void ept_access_violation(unsigned long access, enum ept_access_op op, 2475 u64 expected_qual) 2476 { 2477 ept_violation(EPT_PRESENT, access, op, 2478 expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR); 2479 } 2480 2481 /* 2482 * For translations that don't involve a GVA, that is physical address (paddr) 2483 * accesses, EPT violations don't set the flag EPT_VLT_PADDR. For a typical 2484 * guest memory access, the hardware does GVA -> GPA -> HPA. However, certain 2485 * translations don't involve GVAs, such as when the hardware does the guest 2486 * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU 2487 * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides 2488 * on isn't present in the EPT, then the EPT violation will be for GPA_2 and 2489 * the EPT_VLT_PADDR bit will be clear in the exit qualification. 2490 * 2491 * Note that paddr violations can also be triggered by loading PAE page tables 2492 * with wonky addresses. We don't test that yet. 2493 * 2494 * This function modifies the EPT entry that maps the GPA that the guest page 2495 * table entry mapping ept_access_test_data.gva resides on. 2496 * 2497 * @ept_access EPT permissions to set. Other permissions are cleared. 2498 * 2499 * @pte_ad Set the A/D bits on the guest PTE accordingly. 2500 * 2501 * @op Guest operation to perform with 2502 * ept_access_test_data.gva. 2503 * 2504 * @expect_violation 2505 * Is a violation expected during the paddr access? 2506 * 2507 * @expected_qual Expected qualification for the EPT violation. 2508 * EPT_VLT_PADDR should be clear. 2509 */ 2510 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad, 2511 enum ept_access_op op, bool expect_violation, 2512 u64 expected_qual) 2513 { 2514 struct ept_access_test_data *data = &ept_access_test_data; 2515 unsigned long *ptep; 2516 unsigned long gpa; 2517 unsigned long orig_epte; 2518 2519 /* Modify the guest PTE mapping data->gva according to @pte_ad. */ 2520 ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1); 2521 TEST_ASSERT(ptep); 2522 TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa); 2523 *ptep = (*ptep & ~PT_AD_MASK) | pte_ad; 2524 ept_access_test_guest_flush_tlb(); 2525 2526 /* 2527 * Now modify the access bits on the EPT entry for the GPA that the 2528 * guest PTE resides on. Note that by modifying a single EPT entry, 2529 * we're potentially affecting 512 guest PTEs. However, we've carefully 2530 * constructed our test such that those other 511 PTEs aren't used by 2531 * the guest: data->gva is at the beginning of a 1G huge page, thus the 2532 * PTE we're modifying is at the beginning of a 4K page and the 2533 * following 511 entires are also under our control (and not touched by 2534 * the guest). 2535 */ 2536 gpa = virt_to_phys(ptep); 2537 TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0); 2538 /* 2539 * Make sure the guest page table page is mapped with a 4K EPT entry, 2540 * otherwise our level=1 twiddling below will fail. We use the 2541 * identity map (gpa = gpa) since page tables are shared with the host. 2542 */ 2543 install_ept(pml4, gpa, gpa, EPT_PRESENT); 2544 orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1, 2545 /*clear=*/EPT_PRESENT, /*set=*/ept_access); 2546 2547 if (expect_violation) { 2548 do_ept_violation(/*leaf=*/true, op, 2549 expected_qual | EPT_VLT_LADDR_VLD, gpa); 2550 ept_untwiddle(gpa, /*level=*/1, orig_epte); 2551 do_ept_access_op(op); 2552 } else { 2553 do_ept_access_op(op); 2554 ept_untwiddle(gpa, /*level=*/1, orig_epte); 2555 } 2556 2557 TEST_ASSERT(*ptep & PT_ACCESSED_MASK); 2558 if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE) 2559 TEST_ASSERT(*ptep & PT_DIRTY_MASK); 2560 2561 skip_exit_vmcall(); 2562 } 2563 2564 static void ept_access_allowed_paddr(unsigned long ept_access, 2565 unsigned long pte_ad, 2566 enum ept_access_op op) 2567 { 2568 ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false, 2569 /*expected_qual=*/-1); 2570 } 2571 2572 static void ept_access_violation_paddr(unsigned long ept_access, 2573 unsigned long pte_ad, 2574 enum ept_access_op op, 2575 u64 expected_qual) 2576 { 2577 ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true, 2578 expected_qual); 2579 } 2580 2581 2582 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level, 2583 unsigned long clear, 2584 unsigned long set, 2585 enum ept_access_op op) 2586 { 2587 struct ept_access_test_data *data = &ept_access_test_data; 2588 unsigned long orig_pte; 2589 2590 orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set); 2591 2592 /* No violation. Should proceed to vmcall. */ 2593 do_ept_access_op(op); 2594 skip_exit_vmcall(); 2595 2596 ept_untwiddle(data->gpa, level, orig_pte); 2597 } 2598 2599 static void ept_allowed_at_level(int level, unsigned long clear, 2600 unsigned long set, enum ept_access_op op) 2601 { 2602 ept_allowed_at_level_mkhuge(false, level, clear, set, op); 2603 if (ept_huge_pages_supported(level)) 2604 ept_allowed_at_level_mkhuge(true, level, clear, set, op); 2605 } 2606 2607 static void ept_allowed(unsigned long clear, unsigned long set, 2608 enum ept_access_op op) 2609 { 2610 ept_allowed_at_level(1, clear, set, op); 2611 ept_allowed_at_level(2, clear, set, op); 2612 ept_allowed_at_level(3, clear, set, op); 2613 ept_allowed_at_level(4, clear, set, op); 2614 } 2615 2616 static void ept_ignored_bit(int bit) 2617 { 2618 /* Set the bit. */ 2619 ept_allowed(0, 1ul << bit, OP_READ); 2620 ept_allowed(0, 1ul << bit, OP_WRITE); 2621 ept_allowed(0, 1ul << bit, OP_EXEC); 2622 2623 /* Clear the bit. */ 2624 ept_allowed(1ul << bit, 0, OP_READ); 2625 ept_allowed(1ul << bit, 0, OP_WRITE); 2626 ept_allowed(1ul << bit, 0, OP_EXEC); 2627 } 2628 2629 static void ept_access_allowed(unsigned long access, enum ept_access_op op) 2630 { 2631 ept_allowed(EPT_PRESENT, access, op); 2632 } 2633 2634 2635 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level, 2636 unsigned long clear, 2637 unsigned long set, 2638 enum ept_access_op op) 2639 { 2640 struct ept_access_test_data *data = &ept_access_test_data; 2641 unsigned long orig_pte; 2642 2643 orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set); 2644 2645 do_ept_access_op(op); 2646 assert_exit_reason(VMX_EPT_MISCONFIG); 2647 2648 /* Intel 27.2.1, "For all other VM exits, this field is cleared." */ 2649 #if 0 2650 /* broken: */ 2651 TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0); 2652 #endif 2653 #if 0 2654 /* 2655 * broken: 2656 * According to description of exit qual for EPT violation, 2657 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid. 2658 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought 2659 * to be set for msiconfig. 2660 */ 2661 TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS), 2662 (unsigned long) ( 2663 op == OP_EXEC ? data->gva + 1 : data->gva)); 2664 #endif 2665 2666 /* Fix the violation and resume the op loop. */ 2667 ept_untwiddle(data->gpa, level, orig_pte); 2668 enter_guest(); 2669 skip_exit_vmcall(); 2670 } 2671 2672 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level, 2673 unsigned long clear, 2674 unsigned long set) 2675 { 2676 /* The op shouldn't matter (read, write, exec), so try them all! */ 2677 ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ); 2678 ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE); 2679 ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC); 2680 } 2681 2682 static void ept_misconfig_at_level(int level, unsigned long clear, 2683 unsigned long set) 2684 { 2685 ept_misconfig_at_level_mkhuge(false, level, clear, set); 2686 if (ept_huge_pages_supported(level)) 2687 ept_misconfig_at_level_mkhuge(true, level, clear, set); 2688 } 2689 2690 static void ept_misconfig(unsigned long clear, unsigned long set) 2691 { 2692 ept_misconfig_at_level(1, clear, set); 2693 ept_misconfig_at_level(2, clear, set); 2694 ept_misconfig_at_level(3, clear, set); 2695 ept_misconfig_at_level(4, clear, set); 2696 } 2697 2698 static void ept_access_misconfig(unsigned long access) 2699 { 2700 ept_misconfig(EPT_PRESENT, access); 2701 } 2702 2703 static void ept_reserved_bit_at_level_nohuge(int level, int bit) 2704 { 2705 /* Setting the bit causes a misconfig. */ 2706 ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit); 2707 2708 /* Making the entry non-present turns reserved bits into ignored. */ 2709 ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ, 2710 EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR); 2711 } 2712 2713 static void ept_reserved_bit_at_level_huge(int level, int bit) 2714 { 2715 /* Setting the bit causes a misconfig. */ 2716 ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit); 2717 2718 /* Making the entry non-present turns reserved bits into ignored. */ 2719 ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ, 2720 EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR); 2721 } 2722 2723 static void ept_reserved_bit_at_level(int level, int bit) 2724 { 2725 /* Setting the bit causes a misconfig. */ 2726 ept_misconfig_at_level(level, 0, 1ul << bit); 2727 2728 /* Making the entry non-present turns reserved bits into ignored. */ 2729 ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ, 2730 EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR); 2731 } 2732 2733 static void ept_reserved_bit(int bit) 2734 { 2735 ept_reserved_bit_at_level(1, bit); 2736 ept_reserved_bit_at_level(2, bit); 2737 ept_reserved_bit_at_level(3, bit); 2738 ept_reserved_bit_at_level(4, bit); 2739 } 2740 2741 #define PAGE_2M_ORDER 9 2742 #define PAGE_1G_ORDER 18 2743 2744 static void *get_1g_page(void) 2745 { 2746 static void *alloc; 2747 2748 if (!alloc) 2749 alloc = alloc_pages(PAGE_1G_ORDER); 2750 return alloc; 2751 } 2752 2753 static void ept_access_test_teardown(void *unused) 2754 { 2755 /* Exit the guest cleanly. */ 2756 do_ept_access_op(OP_EXIT); 2757 } 2758 2759 static void ept_access_test_guest(void) 2760 { 2761 struct ept_access_test_data *data = &ept_access_test_data; 2762 int (*code)(void) = (int (*)(void)) &data->gva[1]; 2763 2764 while (true) { 2765 switch (data->op) { 2766 case OP_READ: 2767 TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1); 2768 break; 2769 case OP_WRITE: 2770 *data->gva = MAGIC_VAL_2; 2771 TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2); 2772 *data->gva = MAGIC_VAL_1; 2773 break; 2774 case OP_EXEC: 2775 TEST_ASSERT_EQ(42, code()); 2776 break; 2777 case OP_FLUSH_TLB: 2778 write_cr3(read_cr3()); 2779 break; 2780 case OP_EXIT: 2781 return; 2782 default: 2783 TEST_ASSERT_MSG(false, "Unknown op %d", data->op); 2784 } 2785 vmcall(); 2786 } 2787 } 2788 2789 static void ept_access_test_setup(void) 2790 { 2791 struct ept_access_test_data *data = &ept_access_test_data; 2792 unsigned long npages = 1ul << PAGE_1G_ORDER; 2793 unsigned long size = npages * PAGE_SIZE; 2794 unsigned long *page_table = current_page_table(); 2795 unsigned long pte; 2796 2797 if (setup_ept(false)) 2798 test_skip("EPT not supported"); 2799 2800 /* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */ 2801 if (cpuid_maxphyaddr() < 40) 2802 test_skip("Test needs MAXPHYADDR >= 40"); 2803 2804 test_set_guest(ept_access_test_guest); 2805 test_add_teardown(ept_access_test_teardown, NULL); 2806 2807 data->hva = get_1g_page(); 2808 TEST_ASSERT(data->hva); 2809 data->hpa = virt_to_phys(data->hva); 2810 2811 data->gpa = 1ul << 39; 2812 data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2), 2813 size); 2814 TEST_ASSERT(!any_present_pages(page_table, data->gva, size)); 2815 install_pages(page_table, data->gpa, size, data->gva); 2816 2817 /* 2818 * Make sure nothing's mapped here so the tests that screw with the 2819 * pml4 entry don't inadvertently break something. 2820 */ 2821 TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0); 2822 TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0); 2823 install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT); 2824 2825 data->hva[0] = MAGIC_VAL_1; 2826 memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start); 2827 } 2828 2829 static void ept_access_test_not_present(void) 2830 { 2831 ept_access_test_setup(); 2832 /* --- */ 2833 ept_access_violation(0, OP_READ, EPT_VLT_RD); 2834 ept_access_violation(0, OP_WRITE, EPT_VLT_WR); 2835 ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH); 2836 } 2837 2838 static void ept_access_test_read_only(void) 2839 { 2840 ept_access_test_setup(); 2841 2842 /* r-- */ 2843 ept_access_allowed(EPT_RA, OP_READ); 2844 ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD); 2845 ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD); 2846 } 2847 2848 static void ept_access_test_write_only(void) 2849 { 2850 ept_access_test_setup(); 2851 /* -w- */ 2852 ept_access_misconfig(EPT_WA); 2853 } 2854 2855 static void ept_access_test_read_write(void) 2856 { 2857 ept_access_test_setup(); 2858 /* rw- */ 2859 ept_access_allowed(EPT_RA | EPT_WA, OP_READ); 2860 ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE); 2861 ept_access_violation(EPT_RA | EPT_WA, OP_EXEC, 2862 EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR); 2863 } 2864 2865 2866 static void ept_access_test_execute_only(void) 2867 { 2868 ept_access_test_setup(); 2869 /* --x */ 2870 if (ept_execute_only_supported()) { 2871 ept_access_violation(EPT_EA, OP_READ, 2872 EPT_VLT_RD | EPT_VLT_PERM_EX); 2873 ept_access_violation(EPT_EA, OP_WRITE, 2874 EPT_VLT_WR | EPT_VLT_PERM_EX); 2875 ept_access_allowed(EPT_EA, OP_EXEC); 2876 } else { 2877 ept_access_misconfig(EPT_EA); 2878 } 2879 } 2880 2881 static void ept_access_test_read_execute(void) 2882 { 2883 ept_access_test_setup(); 2884 /* r-x */ 2885 ept_access_allowed(EPT_RA | EPT_EA, OP_READ); 2886 ept_access_violation(EPT_RA | EPT_EA, OP_WRITE, 2887 EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX); 2888 ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC); 2889 } 2890 2891 static void ept_access_test_write_execute(void) 2892 { 2893 ept_access_test_setup(); 2894 /* -wx */ 2895 ept_access_misconfig(EPT_WA | EPT_EA); 2896 } 2897 2898 static void ept_access_test_read_write_execute(void) 2899 { 2900 ept_access_test_setup(); 2901 /* rwx */ 2902 ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ); 2903 ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE); 2904 ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC); 2905 } 2906 2907 static void ept_access_test_reserved_bits(void) 2908 { 2909 int i; 2910 int maxphyaddr; 2911 2912 ept_access_test_setup(); 2913 2914 /* Reserved bits above maxphyaddr. */ 2915 maxphyaddr = cpuid_maxphyaddr(); 2916 for (i = maxphyaddr; i <= 51; i++) { 2917 report_prefix_pushf("reserved_bit=%d", i); 2918 ept_reserved_bit(i); 2919 report_prefix_pop(); 2920 } 2921 2922 /* Level-specific reserved bits. */ 2923 ept_reserved_bit_at_level_nohuge(2, 3); 2924 ept_reserved_bit_at_level_nohuge(2, 4); 2925 ept_reserved_bit_at_level_nohuge(2, 5); 2926 ept_reserved_bit_at_level_nohuge(2, 6); 2927 /* 2M alignment. */ 2928 for (i = 12; i < 20; i++) { 2929 report_prefix_pushf("reserved_bit=%d", i); 2930 ept_reserved_bit_at_level_huge(2, i); 2931 report_prefix_pop(); 2932 } 2933 ept_reserved_bit_at_level_nohuge(3, 3); 2934 ept_reserved_bit_at_level_nohuge(3, 4); 2935 ept_reserved_bit_at_level_nohuge(3, 5); 2936 ept_reserved_bit_at_level_nohuge(3, 6); 2937 /* 1G alignment. */ 2938 for (i = 12; i < 29; i++) { 2939 report_prefix_pushf("reserved_bit=%d", i); 2940 ept_reserved_bit_at_level_huge(3, i); 2941 report_prefix_pop(); 2942 } 2943 ept_reserved_bit_at_level(4, 3); 2944 ept_reserved_bit_at_level(4, 4); 2945 ept_reserved_bit_at_level(4, 5); 2946 ept_reserved_bit_at_level(4, 6); 2947 ept_reserved_bit_at_level(4, 7); 2948 } 2949 2950 static void ept_access_test_ignored_bits(void) 2951 { 2952 ept_access_test_setup(); 2953 /* 2954 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as 2955 * far as translation is concerned even if AD bits are enabled in the 2956 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution 2957 * control is 0. 2958 */ 2959 ept_ignored_bit(8); 2960 ept_ignored_bit(9); 2961 ept_ignored_bit(10); 2962 ept_ignored_bit(11); 2963 ept_ignored_bit(52); 2964 ept_ignored_bit(53); 2965 ept_ignored_bit(54); 2966 ept_ignored_bit(55); 2967 ept_ignored_bit(56); 2968 ept_ignored_bit(57); 2969 ept_ignored_bit(58); 2970 ept_ignored_bit(59); 2971 ept_ignored_bit(60); 2972 ept_ignored_bit(61); 2973 ept_ignored_bit(62); 2974 ept_ignored_bit(63); 2975 } 2976 2977 static void ept_access_test_paddr_not_present_ad_disabled(void) 2978 { 2979 ept_access_test_setup(); 2980 ept_disable_ad_bits(); 2981 2982 ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD); 2983 ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD); 2984 ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD); 2985 } 2986 2987 static void ept_access_test_paddr_not_present_ad_enabled(void) 2988 { 2989 u64 qual = EPT_VLT_RD | EPT_VLT_WR; 2990 2991 ept_access_test_setup(); 2992 ept_enable_ad_bits_or_skip_test(); 2993 2994 ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual); 2995 ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual); 2996 ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual); 2997 } 2998 2999 static void ept_access_test_paddr_read_only_ad_disabled(void) 3000 { 3001 /* 3002 * When EPT AD bits are disabled, all accesses to guest paging 3003 * structures are reported separately as a read and (after 3004 * translation of the GPA to host physical address) a read+write 3005 * if the A/D bits have to be set. 3006 */ 3007 u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD; 3008 3009 ept_access_test_setup(); 3010 ept_disable_ad_bits(); 3011 3012 /* Can't update A bit, so all accesses fail. */ 3013 ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual); 3014 ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual); 3015 ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual); 3016 /* AD bits disabled, so only writes try to update the D bit. */ 3017 ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ); 3018 ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual); 3019 ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC); 3020 /* Both A and D already set, so read-only is OK. */ 3021 ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ); 3022 ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE); 3023 ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC); 3024 } 3025 3026 static void ept_access_test_paddr_read_only_ad_enabled(void) 3027 { 3028 /* 3029 * When EPT AD bits are enabled, all accesses to guest paging 3030 * structures are considered writes as far as EPT translation 3031 * is concerned. 3032 */ 3033 u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD; 3034 3035 ept_access_test_setup(); 3036 ept_enable_ad_bits_or_skip_test(); 3037 3038 ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual); 3039 ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual); 3040 ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual); 3041 ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual); 3042 ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual); 3043 ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual); 3044 ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual); 3045 ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual); 3046 ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual); 3047 } 3048 3049 static void ept_access_test_paddr_read_write(void) 3050 { 3051 ept_access_test_setup(); 3052 /* Read-write access to paging structure. */ 3053 ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ); 3054 ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE); 3055 ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC); 3056 } 3057 3058 static void ept_access_test_paddr_read_write_execute(void) 3059 { 3060 ept_access_test_setup(); 3061 /* RWX access to paging structure. */ 3062 ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ); 3063 ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE); 3064 ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC); 3065 } 3066 3067 static void ept_access_test_paddr_read_execute_ad_disabled(void) 3068 { 3069 /* 3070 * When EPT AD bits are disabled, all accesses to guest paging 3071 * structures are reported separately as a read and (after 3072 * translation of the GPA to host physical address) a read+write 3073 * if the A/D bits have to be set. 3074 */ 3075 u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX; 3076 3077 ept_access_test_setup(); 3078 ept_disable_ad_bits(); 3079 3080 /* Can't update A bit, so all accesses fail. */ 3081 ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual); 3082 ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual); 3083 ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual); 3084 /* AD bits disabled, so only writes try to update the D bit. */ 3085 ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ); 3086 ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual); 3087 ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC); 3088 /* Both A and D already set, so read-only is OK. */ 3089 ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ); 3090 ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE); 3091 ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC); 3092 } 3093 3094 static void ept_access_test_paddr_read_execute_ad_enabled(void) 3095 { 3096 /* 3097 * When EPT AD bits are enabled, all accesses to guest paging 3098 * structures are considered writes as far as EPT translation 3099 * is concerned. 3100 */ 3101 u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX; 3102 3103 ept_access_test_setup(); 3104 ept_enable_ad_bits_or_skip_test(); 3105 3106 ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual); 3107 ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual); 3108 ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual); 3109 ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual); 3110 ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual); 3111 ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual); 3112 ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual); 3113 ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual); 3114 ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual); 3115 } 3116 3117 static void ept_access_test_paddr_not_present_page_fault(void) 3118 { 3119 ept_access_test_setup(); 3120 /* 3121 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is 3122 * page is read-only in EPT but GVA is also mapped read only in PT. 3123 * Thus guest page fault before host takes EPT violation for trying to 3124 * update A bit. 3125 */ 3126 } 3127 3128 static void ept_access_test_force_2m_page(void) 3129 { 3130 ept_access_test_setup(); 3131 3132 TEST_ASSERT_EQ(ept_2m_supported(), true); 3133 ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ); 3134 ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE, 3135 EPT_VLT_WR | EPT_VLT_PERM_RD | 3136 EPT_VLT_LADDR_VLD | EPT_VLT_PADDR); 3137 ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA); 3138 } 3139 3140 static bool invvpid_valid(u64 type, u64 vpid, u64 gla) 3141 { 3142 u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP); 3143 3144 TEST_ASSERT(msr & VPID_CAP_INVVPID); 3145 3146 if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL) 3147 return false; 3148 3149 if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT)))) 3150 return false; 3151 3152 if (vpid >> 16) 3153 return false; 3154 3155 if (type != INVVPID_ALL && !vpid) 3156 return false; 3157 3158 if (type == INVVPID_ADDR && !is_canonical(gla)) 3159 return false; 3160 3161 return true; 3162 } 3163 3164 static void try_invvpid(u64 type, u64 vpid, u64 gla) 3165 { 3166 int rc; 3167 bool valid = invvpid_valid(type, vpid, gla); 3168 u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT 3169 : VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID; 3170 /* 3171 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so 3172 * that we can tell if it is updated by INVVPID. 3173 */ 3174 vmcs_read(~0); 3175 rc = invvpid(type, vpid, gla); 3176 report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type, 3177 vpid, gla, 3178 valid ? "passes" : "fails"); 3179 report(vmcs_read(VMX_INST_ERROR) == expected, 3180 "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)", 3181 rc ? "failed" : "successful", 3182 expected, vmcs_read(VMX_INST_ERROR)); 3183 } 3184 3185 static void ds_invvpid(void *data) 3186 { 3187 u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP); 3188 u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1; 3189 3190 TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL); 3191 asm volatile("invvpid %0, %1" 3192 : 3193 : "m"(*(struct invvpid_operand *)data), 3194 "r"(type)); 3195 } 3196 3197 /* 3198 * The SS override is ignored in 64-bit mode, so we use an addressing 3199 * mode with %rsp as the base register to generate an implicit SS 3200 * reference. 3201 */ 3202 static void ss_invvpid(void *data) 3203 { 3204 u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP); 3205 u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1; 3206 3207 TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL); 3208 asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1" 3209 : "+r"(data) 3210 : "r"(type)); 3211 } 3212 3213 static void invvpid_test_gp(void) 3214 { 3215 bool fault; 3216 3217 fault = test_for_exception(GP_VECTOR, &ds_invvpid, 3218 (void *)NONCANONICAL); 3219 report(fault, "INVVPID with non-canonical DS operand raises #GP"); 3220 } 3221 3222 static void invvpid_test_ss(void) 3223 { 3224 bool fault; 3225 3226 fault = test_for_exception(SS_VECTOR, &ss_invvpid, 3227 (void *)NONCANONICAL); 3228 report(fault, "INVVPID with non-canonical SS operand raises #SS"); 3229 } 3230 3231 static void invvpid_test_pf(void) 3232 { 3233 void *vpage = alloc_vpage(); 3234 bool fault; 3235 3236 fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage); 3237 report(fault, "INVVPID with unmapped operand raises #PF"); 3238 } 3239 3240 static void try_compat_invvpid(void *unused) 3241 { 3242 struct far_pointer32 fp = { 3243 .offset = (uintptr_t)&&invvpid, 3244 .selector = KERNEL_CS32, 3245 }; 3246 register uintptr_t rsp asm("rsp"); 3247 3248 TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid, 3249 "Code address too high."); 3250 TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high."); 3251 3252 asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid); 3253 return; 3254 invvpid: 3255 asm volatile (".code32;" 3256 "invvpid (%eax), %eax;" 3257 "lret;" 3258 ".code64"); 3259 __builtin_unreachable(); 3260 } 3261 3262 static void invvpid_test_compatibility_mode(void) 3263 { 3264 bool fault; 3265 3266 fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL); 3267 report(fault, "Compatibility mode INVVPID raises #UD"); 3268 } 3269 3270 static void invvpid_test_not_in_vmx_operation(void) 3271 { 3272 bool fault; 3273 3274 TEST_ASSERT(!vmx_off()); 3275 fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL); 3276 report(fault, "INVVPID outside of VMX operation raises #UD"); 3277 TEST_ASSERT(!vmx_on()); 3278 } 3279 3280 /* 3281 * This does not test real-address mode, virtual-8086 mode, protected mode, 3282 * or CPL > 0. 3283 */ 3284 static void invvpid_test_v2(void) 3285 { 3286 u64 msr; 3287 int i; 3288 unsigned types = 0; 3289 unsigned type; 3290 3291 if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) || 3292 !(ctrl_cpu_rev[1].clr & CPU_VPID)) 3293 test_skip("VPID not supported"); 3294 3295 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP); 3296 3297 if (!(msr & VPID_CAP_INVVPID)) 3298 test_skip("INVVPID not supported.\n"); 3299 3300 if (msr & VPID_CAP_INVVPID_ADDR) 3301 types |= 1u << INVVPID_ADDR; 3302 if (msr & VPID_CAP_INVVPID_CXTGLB) 3303 types |= 1u << INVVPID_CONTEXT_GLOBAL; 3304 if (msr & VPID_CAP_INVVPID_ALL) 3305 types |= 1u << INVVPID_ALL; 3306 if (msr & VPID_CAP_INVVPID_CXTLOC) 3307 types |= 1u << INVVPID_CONTEXT_LOCAL; 3308 3309 if (!types) 3310 test_skip("No INVVPID types supported.\n"); 3311 3312 for (i = -127; i < 128; i++) 3313 try_invvpid(i, 0xffff, 0); 3314 3315 /* 3316 * VPID must not be more than 16 bits. 3317 */ 3318 for (i = 0; i < 64; i++) 3319 for (type = 0; type < 4; type++) 3320 if (types & (1u << type)) 3321 try_invvpid(type, 1ul << i, 0); 3322 3323 /* 3324 * VPID must not be zero, except for "all contexts." 3325 */ 3326 for (type = 0; type < 4; type++) 3327 if (types & (1u << type)) 3328 try_invvpid(type, 0, 0); 3329 3330 /* 3331 * The gla operand is only validated for single-address INVVPID. 3332 */ 3333 if (types & (1u << INVVPID_ADDR)) 3334 try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL); 3335 3336 invvpid_test_gp(); 3337 invvpid_test_ss(); 3338 invvpid_test_pf(); 3339 invvpid_test_compatibility_mode(); 3340 invvpid_test_not_in_vmx_operation(); 3341 } 3342 3343 /* 3344 * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it 3345 * at least as far as the guest-state checks. Returns false if the 3346 * VMLAUNCH fails early and execution falls through to the next 3347 * instruction. 3348 */ 3349 static bool vmlaunch_succeeds(void) 3350 { 3351 u32 exit_reason; 3352 3353 /* 3354 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to 3355 * unsupported VMCS component"). The caller can then check 3356 * to see if a failed VM-entry sets VMX_INST_ERR as expected. 3357 */ 3358 vmcs_write(~0u, 0); 3359 3360 vmcs_write(HOST_RIP, (uintptr_t)&&success); 3361 __asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch" 3362 : 3363 : "r" ((u64)HOST_RSP) 3364 : "cc", "memory" 3365 : success); 3366 return false; 3367 success: 3368 exit_reason = vmcs_read(EXI_REASON); 3369 TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) || 3370 exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE)); 3371 return true; 3372 } 3373 3374 /* 3375 * Try to launch the current VMCS. 3376 */ 3377 static void test_vmx_vmlaunch(u32 xerror) 3378 { 3379 bool success = vmlaunch_succeeds(); 3380 u32 vmx_inst_err; 3381 3382 report(success == !xerror, "vmlaunch %s", 3383 !xerror ? "succeeds" : "fails"); 3384 if (!success && xerror) { 3385 vmx_inst_err = vmcs_read(VMX_INST_ERROR); 3386 report(vmx_inst_err == xerror, 3387 "VMX inst error is %d (actual %d)", xerror, 3388 vmx_inst_err); 3389 } 3390 } 3391 3392 static void test_vmx_invalid_controls(void) 3393 { 3394 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD); 3395 } 3396 3397 static void test_vmx_valid_controls(void) 3398 { 3399 test_vmx_vmlaunch(0); 3400 } 3401 3402 /* 3403 * Test a particular value of a VM-execution control bit, if the value 3404 * is required or if the value is zero. 3405 */ 3406 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr, 3407 enum Encoding encoding, unsigned bit, 3408 unsigned val) 3409 { 3410 u32 mask = 1u << bit; 3411 bool expected; 3412 u32 controls; 3413 3414 if (msr.set & mask) 3415 TEST_ASSERT(msr.clr & mask); 3416 3417 /* 3418 * We can't arbitrarily turn on a control bit, because it may 3419 * introduce dependencies on other VMCS fields. So, we only 3420 * test turning on bits that have a required setting. 3421 */ 3422 if (val && (msr.clr & mask) && !(msr.set & mask)) 3423 return; 3424 3425 report_prefix_pushf("%s %s bit %d", 3426 val ? "Set" : "Clear", name, bit); 3427 3428 controls = vmcs_read(encoding); 3429 if (val) { 3430 vmcs_write(encoding, msr.set | mask); 3431 expected = (msr.clr & mask); 3432 } else { 3433 vmcs_write(encoding, msr.set & ~mask); 3434 expected = !(msr.set & mask); 3435 } 3436 if (expected) 3437 test_vmx_valid_controls(); 3438 else 3439 test_vmx_invalid_controls(); 3440 vmcs_write(encoding, controls); 3441 report_prefix_pop(); 3442 } 3443 3444 /* 3445 * Test reserved values of a VM-execution control bit, based on the 3446 * allowed bit settings from the corresponding VMX capability MSR. 3447 */ 3448 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr, 3449 enum Encoding encoding, unsigned bit) 3450 { 3451 test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0); 3452 test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1); 3453 } 3454 3455 /* 3456 * Reserved bits in the pin-based VM-execution controls must be set 3457 * properly. Software may consult the VMX capability MSRs to determine 3458 * the proper settings. 3459 * [Intel SDM] 3460 */ 3461 static void test_pin_based_ctls(void) 3462 { 3463 unsigned bit; 3464 3465 printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" : 3466 "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val); 3467 for (bit = 0; bit < 32; bit++) 3468 test_rsvd_ctl_bit("pin-based controls", 3469 ctrl_pin_rev, PIN_CONTROLS, bit); 3470 } 3471 3472 /* 3473 * Reserved bits in the primary processor-based VM-execution controls 3474 * must be set properly. Software may consult the VMX capability MSRs 3475 * to determine the proper settings. 3476 * [Intel SDM] 3477 */ 3478 static void test_primary_processor_based_ctls(void) 3479 { 3480 unsigned bit; 3481 3482 printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" : 3483 "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val); 3484 for (bit = 0; bit < 32; bit++) 3485 test_rsvd_ctl_bit("primary processor-based controls", 3486 ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit); 3487 } 3488 3489 /* 3490 * If the "activate secondary controls" primary processor-based 3491 * VM-execution control is 1, reserved bits in the secondary 3492 * processor-based VM-execution controls must be cleared. Software may 3493 * consult the VMX capability MSRs to determine which bits are 3494 * reserved. 3495 * If the "activate secondary controls" primary processor-based 3496 * VM-execution control is 0 (or if the processor does not support the 3497 * 1-setting of that control), no checks are performed on the 3498 * secondary processor-based VM-execution controls. 3499 * [Intel SDM] 3500 */ 3501 static void test_secondary_processor_based_ctls(void) 3502 { 3503 u32 primary; 3504 u32 secondary; 3505 unsigned bit; 3506 3507 if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) 3508 return; 3509 3510 primary = vmcs_read(CPU_EXEC_CTRL0); 3511 secondary = vmcs_read(CPU_EXEC_CTRL1); 3512 3513 vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY); 3514 printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val); 3515 for (bit = 0; bit < 32; bit++) 3516 test_rsvd_ctl_bit("secondary processor-based controls", 3517 ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit); 3518 3519 /* 3520 * When the "activate secondary controls" VM-execution control 3521 * is clear, there are no checks on the secondary controls. 3522 */ 3523 vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY); 3524 vmcs_write(CPU_EXEC_CTRL1, ~0); 3525 report(vmlaunch_succeeds(), 3526 "Secondary processor-based controls ignored"); 3527 vmcs_write(CPU_EXEC_CTRL1, secondary); 3528 vmcs_write(CPU_EXEC_CTRL0, primary); 3529 } 3530 3531 static void try_cr3_target_count(unsigned i, unsigned max) 3532 { 3533 report_prefix_pushf("CR3 target count 0x%x", i); 3534 vmcs_write(CR3_TARGET_COUNT, i); 3535 if (i <= max) 3536 test_vmx_valid_controls(); 3537 else 3538 test_vmx_invalid_controls(); 3539 report_prefix_pop(); 3540 } 3541 3542 /* 3543 * The CR3-target count must not be greater than 4. Future processors 3544 * may support a different number of CR3-target values. Software 3545 * should read the VMX capability MSR IA32_VMX_MISC to determine the 3546 * number of values supported. 3547 * [Intel SDM] 3548 */ 3549 static void test_cr3_targets(void) 3550 { 3551 unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff; 3552 u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT); 3553 unsigned i; 3554 3555 printf("\nSupported CR3 targets: %d\n", supported_targets); 3556 TEST_ASSERT(supported_targets <= 256); 3557 3558 try_cr3_target_count(-1u, supported_targets); 3559 try_cr3_target_count(0x80000000, supported_targets); 3560 try_cr3_target_count(0x7fffffff, supported_targets); 3561 for (i = 0; i <= supported_targets + 1; i++) 3562 try_cr3_target_count(i, supported_targets); 3563 vmcs_write(CR3_TARGET_COUNT, cr3_targets); 3564 } 3565 3566 /* 3567 * Test a particular address setting in the VMCS 3568 */ 3569 static void test_vmcs_addr(const char *name, 3570 enum Encoding encoding, 3571 u64 align, 3572 bool ignored, 3573 bool skip_beyond_mapped_ram, 3574 u64 addr) 3575 { 3576 report_prefix_pushf("%s = %lx", name, addr); 3577 vmcs_write(encoding, addr); 3578 if (skip_beyond_mapped_ram && 3579 addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align && 3580 addr < (1ul << cpuid_maxphyaddr())) 3581 printf("Skipping physical address beyond mapped RAM\n"); 3582 else if (ignored || (IS_ALIGNED(addr, align) && 3583 addr < (1ul << cpuid_maxphyaddr()))) 3584 test_vmx_valid_controls(); 3585 else 3586 test_vmx_invalid_controls(); 3587 report_prefix_pop(); 3588 } 3589 3590 /* 3591 * Test interesting values for a VMCS address 3592 */ 3593 static void test_vmcs_addr_values(const char *name, 3594 enum Encoding encoding, 3595 u64 align, 3596 bool ignored, 3597 bool skip_beyond_mapped_ram, 3598 u32 bit_start, u32 bit_end) 3599 { 3600 unsigned i; 3601 u64 orig_val = vmcs_read(encoding); 3602 3603 for (i = bit_start; i <= bit_end; i++) 3604 test_vmcs_addr(name, encoding, align, ignored, 3605 skip_beyond_mapped_ram, 1ul << i); 3606 3607 test_vmcs_addr(name, encoding, align, ignored, 3608 skip_beyond_mapped_ram, PAGE_SIZE - 1); 3609 test_vmcs_addr(name, encoding, align, ignored, 3610 skip_beyond_mapped_ram, PAGE_SIZE); 3611 test_vmcs_addr(name, encoding, align, ignored, 3612 skip_beyond_mapped_ram, 3613 (1ul << cpuid_maxphyaddr()) - PAGE_SIZE); 3614 test_vmcs_addr(name, encoding, align, ignored, 3615 skip_beyond_mapped_ram, -1ul); 3616 3617 vmcs_write(encoding, orig_val); 3618 } 3619 3620 /* 3621 * Test a physical address reference in the VMCS, when the corresponding 3622 * feature is enabled and when the corresponding feature is disabled. 3623 */ 3624 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field, 3625 const char *field_name, 3626 const char *control_name, u64 align, 3627 bool skip_beyond_mapped_ram, 3628 bool control_primary) 3629 { 3630 u32 primary = vmcs_read(CPU_EXEC_CTRL0); 3631 u32 secondary = vmcs_read(CPU_EXEC_CTRL1); 3632 u64 page_addr; 3633 3634 if (control_primary) { 3635 if (!(ctrl_cpu_rev[0].clr & control_bit)) 3636 return; 3637 } else { 3638 if (!(ctrl_cpu_rev[1].clr & control_bit)) 3639 return; 3640 } 3641 3642 page_addr = vmcs_read(field); 3643 3644 report_prefix_pushf("%s enabled", control_name); 3645 if (control_primary) { 3646 vmcs_write(CPU_EXEC_CTRL0, primary | control_bit); 3647 } else { 3648 vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY); 3649 vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit); 3650 } 3651 3652 test_vmcs_addr_values(field_name, field, align, false, 3653 skip_beyond_mapped_ram, 0, 63); 3654 report_prefix_pop(); 3655 3656 report_prefix_pushf("%s disabled", control_name); 3657 if (control_primary) { 3658 vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit); 3659 } else { 3660 vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY); 3661 vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit); 3662 } 3663 3664 test_vmcs_addr_values(field_name, field, align, true, false, 0, 63); 3665 report_prefix_pop(); 3666 3667 vmcs_write(field, page_addr); 3668 vmcs_write(CPU_EXEC_CTRL0, primary); 3669 vmcs_write(CPU_EXEC_CTRL1, secondary); 3670 } 3671 3672 /* 3673 * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of 3674 * each I/O-bitmap address must be 0. Neither address should set any 3675 * bits beyond the processor's physical-address width. 3676 * [Intel SDM] 3677 */ 3678 static void test_io_bitmaps(void) 3679 { 3680 test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A, 3681 "I/O bitmap A", "Use I/O bitmaps", 3682 PAGE_SIZE, false, true); 3683 test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B, 3684 "I/O bitmap B", "Use I/O bitmaps", 3685 PAGE_SIZE, false, true); 3686 } 3687 3688 /* 3689 * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of 3690 * the MSR-bitmap address must be 0. The address should not set any 3691 * bits beyond the processor's physical-address width. 3692 * [Intel SDM] 3693 */ 3694 static void test_msr_bitmap(void) 3695 { 3696 test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP, 3697 "MSR bitmap", "Use MSR bitmaps", 3698 PAGE_SIZE, false, true); 3699 } 3700 3701 /* 3702 * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC 3703 * address must satisfy the following checks: 3704 * - Bits 11:0 of the address must be 0. 3705 * - The address should not set any bits beyond the processor's 3706 * physical-address width. 3707 * [Intel SDM] 3708 */ 3709 static void test_apic_virt_addr(void) 3710 { 3711 /* 3712 * Ensure the processor will never use the virtual-APIC page, since 3713 * we will point it to invalid RAM. Otherwise KVM is puzzled about 3714 * what we're trying to achieve and fails vmentry. 3715 */ 3716 u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0); 3717 vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE); 3718 test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR, 3719 "virtual-APIC address", "Use TPR shadow", 3720 PAGE_SIZE, false, true); 3721 vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0); 3722 } 3723 3724 /* 3725 * If the "virtualize APIC-accesses" VM-execution control is 1, the 3726 * APIC-access address must satisfy the following checks: 3727 * - Bits 11:0 of the address must be 0. 3728 * - The address should not set any bits beyond the processor's 3729 * physical-address width. 3730 * [Intel SDM] 3731 */ 3732 static void test_apic_access_addr(void) 3733 { 3734 void *apic_access_page = alloc_page(); 3735 3736 vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page)); 3737 3738 test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR, 3739 "APIC-access address", 3740 "virtualize APIC-accesses", PAGE_SIZE, 3741 true, false); 3742 } 3743 3744 static bool set_bit_pattern(u8 mask, u32 *secondary) 3745 { 3746 u8 i; 3747 bool flag = false; 3748 u32 test_bits[3] = { 3749 CPU_VIRT_X2APIC, 3750 CPU_APIC_REG_VIRT, 3751 CPU_VINTD 3752 }; 3753 3754 for (i = 0; i < ARRAY_SIZE(test_bits); i++) { 3755 if ((mask & (1u << i)) && 3756 (ctrl_cpu_rev[1].clr & test_bits[i])) { 3757 *secondary |= test_bits[i]; 3758 flag = true; 3759 } 3760 } 3761 3762 return (flag); 3763 } 3764 3765 /* 3766 * If the "use TPR shadow" VM-execution control is 0, the following 3767 * VM-execution controls must also be 0: 3768 * - virtualize x2APIC mode 3769 * - APIC-register virtualization 3770 * - virtual-interrupt delivery 3771 * [Intel SDM] 3772 * 3773 * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the 3774 * "virtualize APIC accesses" VM-execution control must be 0. 3775 * [Intel SDM] 3776 */ 3777 static void test_apic_virtual_ctls(void) 3778 { 3779 u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0); 3780 u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1); 3781 u32 primary = saved_primary; 3782 u32 secondary = saved_secondary; 3783 bool ctrl = false; 3784 char str[10] = "disabled"; 3785 u8 i = 0, j; 3786 3787 /* 3788 * First test 3789 */ 3790 if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) == 3791 (CPU_SECONDARY | CPU_TPR_SHADOW))) 3792 return; 3793 3794 primary |= CPU_SECONDARY; 3795 primary &= ~CPU_TPR_SHADOW; 3796 vmcs_write(CPU_EXEC_CTRL0, primary); 3797 3798 while (1) { 3799 for (j = 1; j < 8; j++) { 3800 secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD); 3801 if (primary & CPU_TPR_SHADOW) { 3802 ctrl = true; 3803 } else { 3804 if (! set_bit_pattern(j, &secondary)) 3805 ctrl = true; 3806 else 3807 ctrl = false; 3808 } 3809 3810 vmcs_write(CPU_EXEC_CTRL1, secondary); 3811 report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s", 3812 str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled"); 3813 if (ctrl) 3814 test_vmx_valid_controls(); 3815 else 3816 test_vmx_invalid_controls(); 3817 report_prefix_pop(); 3818 } 3819 3820 if (i == 1) 3821 break; 3822 i++; 3823 3824 primary |= CPU_TPR_SHADOW; 3825 vmcs_write(CPU_EXEC_CTRL0, primary); 3826 strcpy(str, "enabled"); 3827 } 3828 3829 /* 3830 * Second test 3831 */ 3832 u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES); 3833 3834 primary = saved_primary; 3835 secondary = saved_secondary; 3836 if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls)) 3837 return; 3838 3839 vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY); 3840 secondary &= ~CPU_VIRT_APIC_ACCESSES; 3841 vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC); 3842 report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled"); 3843 test_vmx_valid_controls(); 3844 report_prefix_pop(); 3845 3846 vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES); 3847 report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled"); 3848 test_vmx_valid_controls(); 3849 report_prefix_pop(); 3850 3851 vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC); 3852 report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled"); 3853 test_vmx_invalid_controls(); 3854 report_prefix_pop(); 3855 3856 vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES); 3857 report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled"); 3858 test_vmx_valid_controls(); 3859 report_prefix_pop(); 3860 3861 vmcs_write(CPU_EXEC_CTRL0, saved_primary); 3862 vmcs_write(CPU_EXEC_CTRL1, saved_secondary); 3863 } 3864 3865 /* 3866 * If the "virtual-interrupt delivery" VM-execution control is 1, the 3867 * "external-interrupt exiting" VM-execution control must be 1. 3868 * [Intel SDM] 3869 */ 3870 static void test_virtual_intr_ctls(void) 3871 { 3872 u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0); 3873 u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1); 3874 u32 saved_pin = vmcs_read(PIN_CONTROLS); 3875 u32 primary = saved_primary; 3876 u32 secondary = saved_secondary; 3877 u32 pin = saved_pin; 3878 3879 if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) && 3880 (ctrl_pin_rev.clr & PIN_EXTINT))) 3881 return; 3882 3883 vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW); 3884 vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD); 3885 vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT); 3886 report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled"); 3887 test_vmx_valid_controls(); 3888 report_prefix_pop(); 3889 3890 vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD); 3891 report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled"); 3892 test_vmx_invalid_controls(); 3893 report_prefix_pop(); 3894 3895 vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT); 3896 report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled"); 3897 test_vmx_valid_controls(); 3898 report_prefix_pop(); 3899 3900 vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT); 3901 report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled"); 3902 test_vmx_invalid_controls(); 3903 report_prefix_pop(); 3904 3905 vmcs_write(CPU_EXEC_CTRL0, saved_primary); 3906 vmcs_write(CPU_EXEC_CTRL1, saved_secondary); 3907 vmcs_write(PIN_CONTROLS, saved_pin); 3908 } 3909 3910 static void test_pi_desc_addr(u64 addr, bool ctrl) 3911 { 3912 vmcs_write(POSTED_INTR_DESC_ADDR, addr); 3913 report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr); 3914 if (ctrl) 3915 test_vmx_valid_controls(); 3916 else 3917 test_vmx_invalid_controls(); 3918 report_prefix_pop(); 3919 } 3920 3921 /* 3922 * If the “process posted interrupts†VM-execution control is 1, the 3923 * following must be true: 3924 * 3925 * - The “virtual-interrupt delivery†VM-execution control is 1. 3926 * - The “acknowledge interrupt on exit†VM-exit control is 1. 3927 * - The posted-interrupt notification vector has a value in the 3928 * - range 0–255 (bits 15:8 are all 0). 3929 * - Bits 5:0 of the posted-interrupt descriptor address are all 0. 3930 * - The posted-interrupt descriptor address does not set any bits 3931 * beyond the processor's physical-address width. 3932 * [Intel SDM] 3933 */ 3934 static void test_posted_intr(void) 3935 { 3936 u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0); 3937 u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1); 3938 u32 saved_pin = vmcs_read(PIN_CONTROLS); 3939 u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS); 3940 u32 primary = saved_primary; 3941 u32 secondary = saved_secondary; 3942 u32 pin = saved_pin; 3943 u32 exit_ctl = exit_ctl_saved; 3944 u16 vec; 3945 int i; 3946 3947 if (!((ctrl_pin_rev.clr & PIN_POST_INTR) && 3948 (ctrl_cpu_rev[1].clr & CPU_VINTD) && 3949 (ctrl_exit_rev.clr & EXI_INTA))) 3950 return; 3951 3952 vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW); 3953 3954 /* 3955 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit 3956 */ 3957 pin |= PIN_POST_INTR; 3958 vmcs_write(PIN_CONTROLS, pin); 3959 secondary &= ~CPU_VINTD; 3960 vmcs_write(CPU_EXEC_CTRL1, secondary); 3961 report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled"); 3962 test_vmx_invalid_controls(); 3963 report_prefix_pop(); 3964 3965 secondary |= CPU_VINTD; 3966 vmcs_write(CPU_EXEC_CTRL1, secondary); 3967 report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled"); 3968 test_vmx_invalid_controls(); 3969 report_prefix_pop(); 3970 3971 exit_ctl &= ~EXI_INTA; 3972 vmcs_write(EXI_CONTROLS, exit_ctl); 3973 report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled"); 3974 test_vmx_invalid_controls(); 3975 report_prefix_pop(); 3976 3977 exit_ctl |= EXI_INTA; 3978 vmcs_write(EXI_CONTROLS, exit_ctl); 3979 report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled"); 3980 test_vmx_valid_controls(); 3981 report_prefix_pop(); 3982 3983 secondary &= ~CPU_VINTD; 3984 vmcs_write(CPU_EXEC_CTRL1, secondary); 3985 report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled"); 3986 test_vmx_invalid_controls(); 3987 report_prefix_pop(); 3988 3989 secondary |= CPU_VINTD; 3990 vmcs_write(CPU_EXEC_CTRL1, secondary); 3991 report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled"); 3992 test_vmx_valid_controls(); 3993 report_prefix_pop(); 3994 3995 /* 3996 * Test posted-interrupt notification vector 3997 */ 3998 for (i = 0; i < 8; i++) { 3999 vec = (1ul << i); 4000 vmcs_write(PINV, vec); 4001 report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec); 4002 test_vmx_valid_controls(); 4003 report_prefix_pop(); 4004 } 4005 for (i = 8; i < 16; i++) { 4006 vec = (1ul << i); 4007 vmcs_write(PINV, vec); 4008 report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec); 4009 test_vmx_invalid_controls(); 4010 report_prefix_pop(); 4011 } 4012 4013 vec &= ~(0xff << 8); 4014 vmcs_write(PINV, vec); 4015 report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec); 4016 test_vmx_valid_controls(); 4017 report_prefix_pop(); 4018 4019 /* 4020 * Test posted-interrupt descriptor addresss 4021 */ 4022 for (i = 0; i < 6; i++) { 4023 test_pi_desc_addr(1ul << i, false); 4024 } 4025 4026 test_pi_desc_addr(0xf0, false); 4027 test_pi_desc_addr(0xff, false); 4028 test_pi_desc_addr(0x0f, false); 4029 test_pi_desc_addr(0x8000, true); 4030 test_pi_desc_addr(0x00, true); 4031 test_pi_desc_addr(0xc000, true); 4032 4033 test_vmcs_addr_values("process-posted interrupts", 4034 POSTED_INTR_DESC_ADDR, 64, 4035 false, false, 0, 63); 4036 4037 vmcs_write(CPU_EXEC_CTRL0, saved_primary); 4038 vmcs_write(CPU_EXEC_CTRL1, saved_secondary); 4039 vmcs_write(PIN_CONTROLS, saved_pin); 4040 } 4041 4042 static void test_apic_ctls(void) 4043 { 4044 test_apic_virt_addr(); 4045 test_apic_access_addr(); 4046 test_apic_virtual_ctls(); 4047 test_virtual_intr_ctls(); 4048 test_posted_intr(); 4049 } 4050 4051 /* 4052 * If the “enable VPID†VM-execution control is 1, the value of the 4053 * of the VPID VM-execution control field must not be 0000H. 4054 * [Intel SDM] 4055 */ 4056 static void test_vpid(void) 4057 { 4058 u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0); 4059 u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1); 4060 u16 vpid = 0x0000; 4061 int i; 4062 4063 if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) && 4064 (ctrl_cpu_rev[1].clr & CPU_VPID))) { 4065 printf("Secondary controls and/or VPID not supported\n"); 4066 return; 4067 } 4068 4069 vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY); 4070 vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID); 4071 vmcs_write(VPID, vpid); 4072 report_prefix_pushf("VPID disabled; VPID value %x", vpid); 4073 test_vmx_valid_controls(); 4074 report_prefix_pop(); 4075 4076 vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID); 4077 report_prefix_pushf("VPID enabled; VPID value %x", vpid); 4078 test_vmx_invalid_controls(); 4079 report_prefix_pop(); 4080 4081 for (i = 0; i < 16; i++) { 4082 vpid = (short)1 << i;; 4083 vmcs_write(VPID, vpid); 4084 report_prefix_pushf("VPID enabled; VPID value %x", vpid); 4085 test_vmx_valid_controls(); 4086 report_prefix_pop(); 4087 } 4088 4089 vmcs_write(CPU_EXEC_CTRL0, saved_primary); 4090 vmcs_write(CPU_EXEC_CTRL1, saved_secondary); 4091 } 4092 4093 static void set_vtpr(unsigned vtpr) 4094 { 4095 *(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr; 4096 } 4097 4098 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr) 4099 { 4100 bool valid = true; 4101 u32 primary = vmcs_read(CPU_EXEC_CTRL0); 4102 u32 secondary = vmcs_read(CPU_EXEC_CTRL1); 4103 4104 if ((primary & CPU_TPR_SHADOW) && 4105 (!(primary & CPU_SECONDARY) || 4106 !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)))) 4107 valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf); 4108 4109 set_vtpr(vtpr); 4110 report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x", 4111 threshold, (vtpr >> 4) & 0xf); 4112 if (valid) 4113 test_vmx_valid_controls(); 4114 else 4115 test_vmx_invalid_controls(); 4116 report_prefix_pop(); 4117 } 4118 4119 static void test_invalid_event_injection(void) 4120 { 4121 u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO); 4122 u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR); 4123 u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN); 4124 u32 primary_save = vmcs_read(CPU_EXEC_CTRL0); 4125 u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1); 4126 u64 guest_cr0_save = vmcs_read(GUEST_CR0); 4127 u32 ent_intr_info_base = INTR_INFO_VALID_MASK; 4128 u32 ent_intr_info, ent_intr_err, ent_intr_len; 4129 u32 cnt; 4130 4131 /* Setup */ 4132 report_prefix_push("invalid event injection"); 4133 vmcs_write(ENT_INTR_ERROR, 0x00000000); 4134 vmcs_write(ENT_INST_LEN, 0x00000001); 4135 4136 /* The field’s interruption type is not set to a reserved value. */ 4137 ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR; 4138 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4139 "RESERVED interruption type invalid [-]", 4140 ent_intr_info); 4141 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4142 test_vmx_invalid_controls(); 4143 report_prefix_pop(); 4144 4145 ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR | 4146 DE_VECTOR; 4147 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4148 "RESERVED interruption type invalid [+]", 4149 ent_intr_info); 4150 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4151 test_vmx_valid_controls(); 4152 report_prefix_pop(); 4153 4154 /* If the interruption type is other event, the vector is 0. */ 4155 ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR; 4156 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4157 "(OTHER EVENT && vector != 0) invalid [-]", 4158 ent_intr_info); 4159 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4160 test_vmx_invalid_controls(); 4161 report_prefix_pop(); 4162 4163 /* If the interruption type is NMI, the vector is 2 (negative case). */ 4164 ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR; 4165 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4166 "(NMI && vector != 2) invalid [-]", ent_intr_info); 4167 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4168 test_vmx_invalid_controls(); 4169 report_prefix_pop(); 4170 4171 /* If the interruption type is NMI, the vector is 2 (positive case). */ 4172 ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR; 4173 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4174 "(NMI && vector == 2) valid [+]", ent_intr_info); 4175 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4176 test_vmx_valid_controls(); 4177 report_prefix_pop(); 4178 4179 /* 4180 * If the interruption type 4181 * is HW exception, the vector is at most 31. 4182 */ 4183 ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20; 4184 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4185 "(HW exception && vector > 31) invalid [-]", 4186 ent_intr_info); 4187 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4188 test_vmx_invalid_controls(); 4189 report_prefix_pop(); 4190 4191 /* 4192 * deliver-error-code is 1 iff either 4193 * (a) the "unrestricted guest" VM-execution control is 0 4194 * (b) CR0.PE is set. 4195 */ 4196 4197 /* Assert that unrestricted guest is disabled or unsupported */ 4198 assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) || 4199 !(secondary_save & CPU_URG)); 4200 4201 ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 4202 GP_VECTOR; 4203 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4204 "error code <-> (!URG || prot_mode) [-]", 4205 ent_intr_info); 4206 vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG); 4207 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4208 test_vmx_invalid_controls(); 4209 report_prefix_pop(); 4210 4211 ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK | 4212 INTR_TYPE_HARD_EXCEPTION | GP_VECTOR; 4213 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4214 "error code <-> (!URG || prot_mode) [+]", 4215 ent_intr_info); 4216 vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG); 4217 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4218 test_vmx_valid_controls(); 4219 report_prefix_pop(); 4220 4221 if (enable_unrestricted_guest()) 4222 goto skip_unrestricted_guest; 4223 4224 ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK | 4225 INTR_TYPE_HARD_EXCEPTION | GP_VECTOR; 4226 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4227 "error code <-> (!URG || prot_mode) [-]", 4228 ent_intr_info); 4229 vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG); 4230 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4231 test_vmx_invalid_controls(); 4232 report_prefix_pop(); 4233 4234 ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 4235 GP_VECTOR; 4236 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4237 "error code <-> (!URG || prot_mode) [-]", 4238 ent_intr_info); 4239 vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE); 4240 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4241 test_vmx_invalid_controls(); 4242 report_prefix_pop(); 4243 4244 vmcs_write(CPU_EXEC_CTRL1, secondary_save); 4245 vmcs_write(CPU_EXEC_CTRL0, primary_save); 4246 4247 skip_unrestricted_guest: 4248 vmcs_write(GUEST_CR0, guest_cr0_save); 4249 4250 /* deliver-error-code is 1 iff the interruption type is HW exception */ 4251 report_prefix_push("error code <-> HW exception"); 4252 for (cnt = 0; cnt < 8; cnt++) { 4253 u32 exception_type_mask = cnt << 8; 4254 u32 deliver_error_code_mask = 4255 exception_type_mask != INTR_TYPE_HARD_EXCEPTION ? 4256 INTR_INFO_DELIVER_CODE_MASK : 0; 4257 4258 ent_intr_info = ent_intr_info_base | deliver_error_code_mask | 4259 exception_type_mask | GP_VECTOR; 4260 report_prefix_pushf("VM-entry intr info=0x%x [-]", 4261 ent_intr_info); 4262 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4263 test_vmx_invalid_controls(); 4264 report_prefix_pop(); 4265 } 4266 report_prefix_pop(); 4267 4268 /* 4269 * deliver-error-code is 1 iff the the vector 4270 * indicates an exception that would normally deliver an error code 4271 */ 4272 report_prefix_push("error code <-> vector delivers error code"); 4273 for (cnt = 0; cnt < 32; cnt++) { 4274 bool has_error_code = false; 4275 u32 deliver_error_code_mask; 4276 4277 switch (cnt) { 4278 case DF_VECTOR: 4279 case TS_VECTOR: 4280 case NP_VECTOR: 4281 case SS_VECTOR: 4282 case GP_VECTOR: 4283 case PF_VECTOR: 4284 case AC_VECTOR: 4285 has_error_code = true; 4286 } 4287 4288 /* Negative case */ 4289 deliver_error_code_mask = has_error_code ? 4290 0 : 4291 INTR_INFO_DELIVER_CODE_MASK; 4292 ent_intr_info = ent_intr_info_base | deliver_error_code_mask | 4293 INTR_TYPE_HARD_EXCEPTION | cnt; 4294 report_prefix_pushf("VM-entry intr info=0x%x [-]", 4295 ent_intr_info); 4296 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4297 test_vmx_invalid_controls(); 4298 report_prefix_pop(); 4299 4300 /* Positive case */ 4301 deliver_error_code_mask = has_error_code ? 4302 INTR_INFO_DELIVER_CODE_MASK : 4303 0; 4304 ent_intr_info = ent_intr_info_base | deliver_error_code_mask | 4305 INTR_TYPE_HARD_EXCEPTION | cnt; 4306 report_prefix_pushf("VM-entry intr info=0x%x [+]", 4307 ent_intr_info); 4308 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4309 test_vmx_valid_controls(); 4310 report_prefix_pop(); 4311 } 4312 report_prefix_pop(); 4313 4314 /* Reserved bits in the field (30:12) are 0. */ 4315 report_prefix_push("reserved bits clear"); 4316 for (cnt = 12; cnt <= 30; cnt++) { 4317 ent_intr_info = ent_intr_info_base | 4318 INTR_INFO_DELIVER_CODE_MASK | 4319 INTR_TYPE_HARD_EXCEPTION | GP_VECTOR | 4320 (1U << cnt); 4321 report_prefix_pushf("VM-entry intr info=0x%x [-]", 4322 ent_intr_info); 4323 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4324 test_vmx_invalid_controls(); 4325 report_prefix_pop(); 4326 } 4327 report_prefix_pop(); 4328 4329 /* 4330 * If deliver-error-code is 1 4331 * bits 31:16 of the VM-entry exception error-code field are 0. 4332 */ 4333 ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK | 4334 INTR_TYPE_HARD_EXCEPTION | GP_VECTOR; 4335 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4336 "VM-entry exception error code[31:16] clear", 4337 ent_intr_info); 4338 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4339 for (cnt = 16; cnt <= 31; cnt++) { 4340 ent_intr_err = 1U << cnt; 4341 report_prefix_pushf("VM-entry intr error=0x%x [-]", 4342 ent_intr_err); 4343 vmcs_write(ENT_INTR_ERROR, ent_intr_err); 4344 test_vmx_invalid_controls(); 4345 report_prefix_pop(); 4346 } 4347 vmcs_write(ENT_INTR_ERROR, 0x00000000); 4348 report_prefix_pop(); 4349 4350 /* 4351 * If the interruption type is software interrupt, software exception, 4352 * or privileged software exception, the VM-entry instruction-length 4353 * field is in the range 0–15. 4354 */ 4355 4356 for (cnt = 0; cnt < 3; cnt++) { 4357 switch (cnt) { 4358 case 0: 4359 ent_intr_info = ent_intr_info_base | 4360 INTR_TYPE_SOFT_INTR; 4361 break; 4362 case 1: 4363 ent_intr_info = ent_intr_info_base | 4364 INTR_TYPE_SOFT_EXCEPTION; 4365 break; 4366 case 2: 4367 ent_intr_info = ent_intr_info_base | 4368 INTR_TYPE_PRIV_SW_EXCEPTION; 4369 break; 4370 } 4371 report_prefix_pushf("%s, VM-entry intr info=0x%x", 4372 "VM-entry instruction-length check", 4373 ent_intr_info); 4374 vmcs_write(ENT_INTR_INFO, ent_intr_info); 4375 4376 /* Instruction length set to -1 (0xFFFFFFFF) should fail */ 4377 ent_intr_len = -1; 4378 report_prefix_pushf("VM-entry intr length = 0x%x [-]", 4379 ent_intr_len); 4380 vmcs_write(ENT_INST_LEN, ent_intr_len); 4381 test_vmx_invalid_controls(); 4382 report_prefix_pop(); 4383 4384 /* Instruction length set to 16 should fail */ 4385 ent_intr_len = 0x00000010; 4386 report_prefix_pushf("VM-entry intr length = 0x%x [-]", 4387 ent_intr_len); 4388 vmcs_write(ENT_INST_LEN, 0x00000010); 4389 test_vmx_invalid_controls(); 4390 report_prefix_pop(); 4391 4392 report_prefix_pop(); 4393 } 4394 4395 /* Cleanup */ 4396 vmcs_write(ENT_INTR_INFO, ent_intr_info_save); 4397 vmcs_write(ENT_INTR_ERROR, ent_intr_error_save); 4398 vmcs_write(ENT_INST_LEN, ent_inst_len_save); 4399 vmcs_write(CPU_EXEC_CTRL0, primary_save); 4400 vmcs_write(CPU_EXEC_CTRL1, secondary_save); 4401 vmcs_write(GUEST_CR0, guest_cr0_save); 4402 report_prefix_pop(); 4403 } 4404 4405 /* 4406 * Test interesting vTPR values for a given TPR threshold. 4407 */ 4408 static void test_vtpr_values(unsigned threshold) 4409 { 4410 try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4); 4411 try_tpr_threshold_and_vtpr(threshold, threshold << 4); 4412 try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4); 4413 } 4414 4415 static void try_tpr_threshold(unsigned threshold) 4416 { 4417 bool valid = true; 4418 4419 u32 primary = vmcs_read(CPU_EXEC_CTRL0); 4420 u32 secondary = vmcs_read(CPU_EXEC_CTRL1); 4421 4422 if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) && 4423 (secondary & CPU_VINTD))) 4424 valid = !(threshold >> 4); 4425 4426 set_vtpr(-1); 4427 vmcs_write(TPR_THRESHOLD, threshold); 4428 report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold); 4429 if (valid) 4430 test_vmx_valid_controls(); 4431 else 4432 test_vmx_invalid_controls(); 4433 report_prefix_pop(); 4434 4435 if (valid) 4436 test_vtpr_values(threshold); 4437 } 4438 4439 /* 4440 * Test interesting TPR threshold values. 4441 */ 4442 static void test_tpr_threshold_values(void) 4443 { 4444 unsigned i; 4445 4446 for (i = 0; i < 0x10; i++) 4447 try_tpr_threshold(i); 4448 for (i = 4; i < 32; i++) 4449 try_tpr_threshold(1u << i); 4450 try_tpr_threshold(-1u); 4451 try_tpr_threshold(0x7fffffff); 4452 } 4453 4454 /* 4455 * This test covers the following two VM entry checks: 4456 * 4457 * i) If the "use TPR shadow" VM-execution control is 1 and the 4458 * "virtual-interrupt delivery" VM-execution control is 0, bits 4459 * 31:4 of the TPR threshold VM-execution control field must 4460 be 0. 4461 * [Intel SDM] 4462 * 4463 * ii) If the "use TPR shadow" VM-execution control is 1, the 4464 * "virtual-interrupt delivery" VM-execution control is 0 4465 * and the "virtualize APIC accesses" VM-execution control 4466 * is 0, the value of bits 3:0 of the TPR threshold VM-execution 4467 * control field must not be greater than the value of bits 4468 * 7:4 of VTPR. 4469 * [Intel SDM] 4470 */ 4471 static void test_tpr_threshold(void) 4472 { 4473 u32 primary = vmcs_read(CPU_EXEC_CTRL0); 4474 u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR); 4475 u64 threshold = vmcs_read(TPR_THRESHOLD); 4476 void *virtual_apic_page; 4477 4478 if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) 4479 return; 4480 4481 virtual_apic_page = alloc_page(); 4482 memset(virtual_apic_page, 0xff, PAGE_SIZE); 4483 vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page)); 4484 4485 vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY)); 4486 report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled"); 4487 test_tpr_threshold_values(); 4488 report_prefix_pop(); 4489 vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW); 4490 report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled"); 4491 test_tpr_threshold_values(); 4492 report_prefix_pop(); 4493 4494 if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) && 4495 (ctrl_cpu_rev[1].clr & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)))) 4496 goto out; 4497 u32 secondary = vmcs_read(CPU_EXEC_CTRL1); 4498 4499 if (ctrl_cpu_rev[1].clr & CPU_VINTD) { 4500 vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD); 4501 report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled"); 4502 test_tpr_threshold_values(); 4503 report_prefix_pop(); 4504 4505 vmcs_write(CPU_EXEC_CTRL0, 4506 vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY); 4507 report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled"); 4508 test_tpr_threshold_values(); 4509 report_prefix_pop(); 4510 } 4511 4512 if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) { 4513 vmcs_write(CPU_EXEC_CTRL0, 4514 vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY); 4515 vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES); 4516 report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled"); 4517 test_tpr_threshold_values(); 4518 report_prefix_pop(); 4519 4520 vmcs_write(CPU_EXEC_CTRL0, 4521 vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY); 4522 report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled"); 4523 test_tpr_threshold_values(); 4524 report_prefix_pop(); 4525 } 4526 4527 if ((ctrl_cpu_rev[1].clr & 4528 (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) == 4529 (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) { 4530 vmcs_write(CPU_EXEC_CTRL0, 4531 vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY); 4532 vmcs_write(CPU_EXEC_CTRL1, 4533 CPU_VINTD | CPU_VIRT_APIC_ACCESSES); 4534 report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled"); 4535 test_tpr_threshold_values(); 4536 report_prefix_pop(); 4537 4538 vmcs_write(CPU_EXEC_CTRL0, 4539 vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY); 4540 report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled"); 4541 test_tpr_threshold_values(); 4542 report_prefix_pop(); 4543 } 4544 4545 vmcs_write(CPU_EXEC_CTRL1, secondary); 4546 out: 4547 vmcs_write(TPR_THRESHOLD, threshold); 4548 vmcs_write(APIC_VIRT_ADDR, apic_virt_addr); 4549 vmcs_write(CPU_EXEC_CTRL0, primary); 4550 } 4551 4552 /* 4553 * This test verifies the following two vmentry checks: 4554 * 4555 * If the "NMI exiting" VM-execution control is 0, "Virtual NMIs" 4556 * VM-execution control must be 0. 4557 * [Intel SDM] 4558 * 4559 * If the “virtual NMIs” VM-execution control is 0, the “NMI-window 4560 * exiting” VM-execution control must be 0. 4561 * [Intel SDM] 4562 */ 4563 static void test_nmi_ctrls(void) 4564 { 4565 u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0; 4566 4567 if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) != 4568 (PIN_NMI | PIN_VIRT_NMI)) { 4569 printf("NMI exiting and Virtual NMIs are not supported !\n"); 4570 return; 4571 } 4572 4573 /* Save the controls so that we can restore them after our tests */ 4574 pin_ctrls = vmcs_read(PIN_CONTROLS); 4575 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0); 4576 4577 test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI); 4578 test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW; 4579 4580 vmcs_write(PIN_CONTROLS, test_pin_ctrls); 4581 report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled"); 4582 test_vmx_valid_controls(); 4583 report_prefix_pop(); 4584 4585 vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI); 4586 report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled"); 4587 test_vmx_invalid_controls(); 4588 report_prefix_pop(); 4589 4590 vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI)); 4591 report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled"); 4592 test_vmx_valid_controls(); 4593 report_prefix_pop(); 4594 4595 vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI); 4596 report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled"); 4597 test_vmx_valid_controls(); 4598 report_prefix_pop(); 4599 4600 if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) { 4601 report_info("NMI-window exiting is not supported, skipping..."); 4602 goto done; 4603 } 4604 4605 vmcs_write(PIN_CONTROLS, test_pin_ctrls); 4606 vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW); 4607 report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled"); 4608 test_vmx_invalid_controls(); 4609 report_prefix_pop(); 4610 4611 vmcs_write(PIN_CONTROLS, test_pin_ctrls); 4612 vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0); 4613 report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled"); 4614 test_vmx_valid_controls(); 4615 report_prefix_pop(); 4616 4617 vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI)); 4618 vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW); 4619 report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled"); 4620 test_vmx_valid_controls(); 4621 report_prefix_pop(); 4622 4623 vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI)); 4624 vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0); 4625 report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled"); 4626 test_vmx_valid_controls(); 4627 report_prefix_pop(); 4628 4629 /* Restore the controls to their original values */ 4630 vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0); 4631 done: 4632 vmcs_write(PIN_CONTROLS, pin_ctrls); 4633 } 4634 4635 static void test_eptp_ad_bit(u64 eptp, bool ctrl) 4636 { 4637 vmcs_write(EPTP, eptp); 4638 report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s", 4639 (eptp & EPTP_AD_FLAG) ? "1": "0"); 4640 if (ctrl) 4641 test_vmx_valid_controls(); 4642 else 4643 test_vmx_invalid_controls(); 4644 report_prefix_pop(); 4645 4646 } 4647 4648 /* 4649 * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution" 4650 * control field must satisfy the following checks: 4651 * 4652 * - The EPT memory type (bits 2:0) must be a value supported by the 4653 * processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR. 4654 * - Bits 5:3 (1 less than the EPT page-walk length) must be 3, 4655 * indicating an EPT page-walk length of 4. 4656 * - Bit 6 (enable bit for accessed and dirty flags for EPT) must be 4657 * 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0, 4658 * indicating that the processor does not support accessed and dirty 4659 * dirty flags for EPT. 4660 * - Reserved bits 11:7 and 63:N (where N is the processor's 4661 * physical-address width) must all be 0. 4662 * 4663 * 2. If the "unrestricted guest" VM-execution control is 1, the 4664 * "enable EPT" VM-execution control must also be 1. 4665 */ 4666 static void test_ept_eptp(void) 4667 { 4668 u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0); 4669 u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1); 4670 u64 eptp_saved = vmcs_read(EPTP); 4671 u32 primary = primary_saved; 4672 u32 secondary = secondary_saved; 4673 u64 msr, eptp = eptp_saved; 4674 bool un_cache = false; 4675 bool wr_bk = false; 4676 bool ctrl; 4677 u32 i, maxphysaddr; 4678 u64 j, resv_bits_mask = 0; 4679 4680 if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) && 4681 (ctrl_cpu_rev[1].clr & CPU_EPT))) { 4682 printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n"); 4683 return; 4684 } 4685 4686 /* 4687 * Memory type (bits 2:0) 4688 */ 4689 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP); 4690 if (msr & EPT_CAP_UC) 4691 un_cache = true; 4692 if (msr & EPT_CAP_WB) 4693 wr_bk = true; 4694 4695 primary |= CPU_SECONDARY; 4696 vmcs_write(CPU_EXEC_CTRL0, primary); 4697 secondary |= CPU_EPT; 4698 vmcs_write(CPU_EXEC_CTRL1, secondary); 4699 eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) | 4700 (3ul << EPTP_PG_WALK_LEN_SHIFT); 4701 vmcs_write(EPTP, eptp); 4702 4703 for (i = 0; i < 8; i++) { 4704 if (i == 0) { 4705 if (un_cache) { 4706 report_info("EPT paging structure memory-type is Un-cacheable\n"); 4707 ctrl = true; 4708 } else { 4709 ctrl = false; 4710 } 4711 } else if (i == 6) { 4712 if (wr_bk) { 4713 report_info("EPT paging structure memory-type is Write-back\n"); 4714 ctrl = true; 4715 } else { 4716 ctrl = false; 4717 } 4718 } else { 4719 ctrl = false; 4720 } 4721 4722 eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i; 4723 vmcs_write(EPTP, eptp); 4724 report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu", 4725 eptp & EPT_MEM_TYPE_MASK); 4726 if (ctrl) 4727 test_vmx_valid_controls(); 4728 else 4729 test_vmx_invalid_controls(); 4730 report_prefix_pop(); 4731 } 4732 4733 eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul; 4734 4735 /* 4736 * Page walk length (bits 5:3) 4737 */ 4738 for (i = 0; i < 8; i++) { 4739 eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) | 4740 (i << EPTP_PG_WALK_LEN_SHIFT); 4741 if (i == 3) 4742 ctrl = true; 4743 else 4744 ctrl = false; 4745 4746 vmcs_write(EPTP, eptp); 4747 report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu", 4748 eptp & EPTP_PG_WALK_LEN_MASK); 4749 if (ctrl) 4750 test_vmx_valid_controls(); 4751 else 4752 test_vmx_invalid_controls(); 4753 report_prefix_pop(); 4754 } 4755 4756 eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) | 4757 3ul << EPTP_PG_WALK_LEN_SHIFT; 4758 4759 /* 4760 * Accessed and dirty flag (bit 6) 4761 */ 4762 if (msr & EPT_CAP_AD_FLAG) { 4763 report_info("Processor supports accessed and dirty flag"); 4764 eptp &= ~EPTP_AD_FLAG; 4765 test_eptp_ad_bit(eptp, true); 4766 4767 eptp |= EPTP_AD_FLAG; 4768 test_eptp_ad_bit(eptp, true); 4769 } else { 4770 report_info("Processor does not supports accessed and dirty flag"); 4771 eptp &= ~EPTP_AD_FLAG; 4772 test_eptp_ad_bit(eptp, true); 4773 4774 eptp |= EPTP_AD_FLAG; 4775 test_eptp_ad_bit(eptp, false); 4776 } 4777 4778 /* 4779 * Reserved bits [11:7] and [63:N] 4780 */ 4781 for (i = 0; i < 32; i++) { 4782 eptp = (eptp & 4783 ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) | 4784 (i << EPTP_RESERV_BITS_SHIFT); 4785 vmcs_write(EPTP, eptp); 4786 report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu", 4787 (eptp >> EPTP_RESERV_BITS_SHIFT) & 4788 EPTP_RESERV_BITS_MASK); 4789 if (i == 0) 4790 test_vmx_valid_controls(); 4791 else 4792 test_vmx_invalid_controls(); 4793 report_prefix_pop(); 4794 } 4795 4796 eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)); 4797 4798 maxphysaddr = cpuid_maxphyaddr(); 4799 for (i = 0; i < (63 - maxphysaddr + 1); i++) { 4800 resv_bits_mask |= 1ul << i; 4801 } 4802 4803 for (j = maxphysaddr - 1; j <= 63; j++) { 4804 eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) | 4805 (j < maxphysaddr ? 0 : 1ul << j); 4806 vmcs_write(EPTP, eptp); 4807 report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu", 4808 (eptp >> maxphysaddr) & resv_bits_mask); 4809 if (j < maxphysaddr) 4810 test_vmx_valid_controls(); 4811 else 4812 test_vmx_invalid_controls(); 4813 report_prefix_pop(); 4814 } 4815 4816 secondary &= ~(CPU_EPT | CPU_URG); 4817 vmcs_write(CPU_EXEC_CTRL1, secondary); 4818 report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled"); 4819 test_vmx_valid_controls(); 4820 report_prefix_pop(); 4821 4822 if (!(ctrl_cpu_rev[1].clr & CPU_URG)) 4823 goto skip_unrestricted_guest; 4824 4825 secondary |= CPU_URG; 4826 vmcs_write(CPU_EXEC_CTRL1, secondary); 4827 report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled"); 4828 test_vmx_invalid_controls(); 4829 report_prefix_pop(); 4830 4831 secondary |= CPU_EPT; 4832 setup_dummy_ept(); 4833 report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled"); 4834 test_vmx_valid_controls(); 4835 report_prefix_pop(); 4836 4837 skip_unrestricted_guest: 4838 secondary &= ~CPU_URG; 4839 vmcs_write(CPU_EXEC_CTRL1, secondary); 4840 report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled"); 4841 test_vmx_valid_controls(); 4842 report_prefix_pop(); 4843 4844 vmcs_write(CPU_EXEC_CTRL0, primary_saved); 4845 vmcs_write(CPU_EXEC_CTRL1, secondary_saved); 4846 vmcs_write(EPTP, eptp_saved); 4847 } 4848 4849 /* 4850 * If the 'enable PML' VM-execution control is 1, the 'enable EPT' 4851 * VM-execution control must also be 1. In addition, the PML address 4852 * must satisfy the following checks: 4853 * 4854 * * Bits 11:0 of the address must be 0. 4855 * * The address should not set any bits beyond the processor's 4856 * physical-address width. 4857 * 4858 * [Intel SDM] 4859 */ 4860 static void test_pml(void) 4861 { 4862 u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0); 4863 u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1); 4864 u32 primary = primary_saved; 4865 u32 secondary = secondary_saved; 4866 4867 if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) && 4868 (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) { 4869 printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n"); 4870 return; 4871 } 4872 4873 primary |= CPU_SECONDARY; 4874 vmcs_write(CPU_EXEC_CTRL0, primary); 4875 secondary &= ~(CPU_PML | CPU_EPT); 4876 vmcs_write(CPU_EXEC_CTRL1, secondary); 4877 report_prefix_pushf("enable-PML disabled, enable-EPT disabled"); 4878 test_vmx_valid_controls(); 4879 report_prefix_pop(); 4880 4881 secondary |= CPU_PML; 4882 vmcs_write(CPU_EXEC_CTRL1, secondary); 4883 report_prefix_pushf("enable-PML enabled, enable-EPT disabled"); 4884 test_vmx_invalid_controls(); 4885 report_prefix_pop(); 4886 4887 secondary |= CPU_EPT; 4888 setup_dummy_ept(); 4889 report_prefix_pushf("enable-PML enabled, enable-EPT enabled"); 4890 test_vmx_valid_controls(); 4891 report_prefix_pop(); 4892 4893 secondary &= ~CPU_PML; 4894 vmcs_write(CPU_EXEC_CTRL1, secondary); 4895 report_prefix_pushf("enable-PML disabled, enable EPT enabled"); 4896 test_vmx_valid_controls(); 4897 report_prefix_pop(); 4898 4899 test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML", 4900 PAGE_SIZE, false, false); 4901 4902 vmcs_write(CPU_EXEC_CTRL0, primary_saved); 4903 vmcs_write(CPU_EXEC_CTRL1, secondary_saved); 4904 } 4905 4906 /* 4907 * If the "activate VMX-preemption timer" VM-execution control is 0, the 4908 * the "save VMX-preemption timer value" VM-exit control must also be 0. 4909 * 4910 * [Intel SDM] 4911 */ 4912 static void test_vmx_preemption_timer(void) 4913 { 4914 u32 saved_pin = vmcs_read(PIN_CONTROLS); 4915 u32 saved_exit = vmcs_read(EXI_CONTROLS); 4916 u32 pin = saved_pin; 4917 u32 exit = saved_exit; 4918 4919 if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) || 4920 (ctrl_pin_rev.clr & PIN_PREEMPT))) { 4921 printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n"); 4922 return; 4923 } 4924 4925 pin |= PIN_PREEMPT; 4926 vmcs_write(PIN_CONTROLS, pin); 4927 exit &= ~EXI_SAVE_PREEMPT; 4928 vmcs_write(EXI_CONTROLS, exit); 4929 report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled"); 4930 test_vmx_valid_controls(); 4931 report_prefix_pop(); 4932 4933 exit |= EXI_SAVE_PREEMPT; 4934 vmcs_write(EXI_CONTROLS, exit); 4935 report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled"); 4936 test_vmx_valid_controls(); 4937 report_prefix_pop(); 4938 4939 pin &= ~PIN_PREEMPT; 4940 vmcs_write(PIN_CONTROLS, pin); 4941 report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled"); 4942 test_vmx_invalid_controls(); 4943 report_prefix_pop(); 4944 4945 exit &= ~EXI_SAVE_PREEMPT; 4946 vmcs_write(EXI_CONTROLS, exit); 4947 report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled"); 4948 test_vmx_valid_controls(); 4949 report_prefix_pop(); 4950 4951 vmcs_write(PIN_CONTROLS, saved_pin); 4952 vmcs_write(EXI_CONTROLS, saved_exit); 4953 } 4954 4955 /* 4956 * Tests for VM-execution control fields 4957 */ 4958 static void test_vm_execution_ctls(void) 4959 { 4960 test_pin_based_ctls(); 4961 test_primary_processor_based_ctls(); 4962 test_secondary_processor_based_ctls(); 4963 test_cr3_targets(); 4964 test_io_bitmaps(); 4965 test_msr_bitmap(); 4966 test_apic_ctls(); 4967 test_tpr_threshold(); 4968 test_nmi_ctrls(); 4969 test_pml(); 4970 test_vpid(); 4971 test_ept_eptp(); 4972 test_vmx_preemption_timer(); 4973 } 4974 4975 /* 4976 * The following checks are performed for the VM-entry MSR-load address if 4977 * the VM-entry MSR-load count field is non-zero: 4978 * 4979 * - The lower 4 bits of the VM-entry MSR-load address must be 0. 4980 * The address should not set any bits beyond the processor’s 4981 * physical-address width. 4982 * 4983 * - The address of the last byte in the VM-entry MSR-load area 4984 * should not set any bits beyond the processor’s physical-address 4985 * width. The address of this last byte is VM-entry MSR-load address 4986 * + (MSR count * 16) - 1. (The arithmetic used for the computation 4987 * uses more bits than the processor’s physical-address width.) 4988 * 4989 * 4990 * [Intel SDM] 4991 */ 4992 static void test_entry_msr_load(void) 4993 { 4994 entry_msr_load = alloc_page(); 4995 u64 tmp; 4996 u32 entry_msr_ld_cnt = 1; 4997 int i; 4998 u32 addr_len = 64; 4999 5000 vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt); 5001 5002 /* Check first 4 bits of VM-entry MSR-load address */ 5003 for (i = 0; i < 4; i++) { 5004 tmp = (u64)entry_msr_load | 1ull << i; 5005 vmcs_write(ENTER_MSR_LD_ADDR, tmp); 5006 report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx", 5007 tmp & 0xf); 5008 test_vmx_invalid_controls(); 5009 report_prefix_pop(); 5010 } 5011 5012 if (basic.val & (1ul << 48)) 5013 addr_len = 32; 5014 5015 test_vmcs_addr_values("VM-entry-MSR-load address", 5016 ENTER_MSR_LD_ADDR, 16, false, false, 5017 4, addr_len - 1); 5018 5019 /* 5020 * Check last byte of VM-entry MSR-load address 5021 */ 5022 entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf); 5023 5024 for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len); 5025 i < 64; i++) { 5026 tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) | 5027 1ul << i; 5028 vmcs_write(ENTER_MSR_LD_ADDR, 5029 tmp - (entry_msr_ld_cnt * 16 - 1)); 5030 test_vmx_invalid_controls(); 5031 } 5032 5033 vmcs_write(ENT_MSR_LD_CNT, 2); 5034 vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16); 5035 test_vmx_invalid_controls(); 5036 vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32); 5037 test_vmx_valid_controls(); 5038 vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48); 5039 test_vmx_valid_controls(); 5040 } 5041 5042 static struct vmx_state_area_test_data { 5043 u32 msr; 5044 u64 exp; 5045 bool enabled; 5046 } vmx_state_area_test_data; 5047 5048 static void guest_state_test_main(void) 5049 { 5050 u64 obs; 5051 struct vmx_state_area_test_data *data = &vmx_state_area_test_data; 5052 5053 while (1) { 5054 if (vmx_get_test_stage() == 2) 5055 break; 5056 5057 if (data->enabled) { 5058 obs = rdmsr(data->msr); 5059 report(data->exp == obs, 5060 "Guest state is 0x%lx (expected 0x%lx)", 5061 obs, data->exp); 5062 } 5063 5064 vmcall(); 5065 } 5066 5067 asm volatile("fnop"); 5068 } 5069 5070 static void advance_guest_state_test(void) 5071 { 5072 u32 reason = vmcs_read(EXI_REASON); 5073 if (! (reason & 0x80000000)) { 5074 u64 guest_rip = vmcs_read(GUEST_RIP); 5075 u32 insn_len = vmcs_read(EXI_INST_LEN); 5076 vmcs_write(GUEST_RIP, guest_rip + insn_len); 5077 } 5078 } 5079 5080 static void report_guest_state_test(const char *test, u32 xreason, 5081 u64 field, const char * field_name) 5082 { 5083 u32 reason = vmcs_read(EXI_REASON); 5084 5085 report(reason == xreason, "%s, %s %lx", test, field_name, field); 5086 advance_guest_state_test(); 5087 } 5088 5089 /* 5090 * Tests for VM-entry control fields 5091 */ 5092 static void test_vm_entry_ctls(void) 5093 { 5094 test_invalid_event_injection(); 5095 test_entry_msr_load(); 5096 } 5097 5098 /* 5099 * The following checks are performed for the VM-exit MSR-store address if 5100 * the VM-exit MSR-store count field is non-zero: 5101 * 5102 * - The lower 4 bits of the VM-exit MSR-store address must be 0. 5103 * The address should not set any bits beyond the processor’s 5104 * physical-address width. 5105 * 5106 * - The address of the last byte in the VM-exit MSR-store area 5107 * should not set any bits beyond the processor’s physical-address 5108 * width. The address of this last byte is VM-exit MSR-store address 5109 * + (MSR count * 16) - 1. (The arithmetic used for the computation 5110 * uses more bits than the processor’s physical-address width.) 5111 * 5112 * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits 5113 * in the range 63:32. 5114 * 5115 * [Intel SDM] 5116 */ 5117 static void test_exit_msr_store(void) 5118 { 5119 exit_msr_store = alloc_page(); 5120 u64 tmp; 5121 u32 exit_msr_st_cnt = 1; 5122 int i; 5123 u32 addr_len = 64; 5124 5125 vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt); 5126 5127 /* Check first 4 bits of VM-exit MSR-store address */ 5128 for (i = 0; i < 4; i++) { 5129 tmp = (u64)exit_msr_store | 1ull << i; 5130 vmcs_write(EXIT_MSR_ST_ADDR, tmp); 5131 report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx", 5132 tmp & 0xf); 5133 test_vmx_invalid_controls(); 5134 report_prefix_pop(); 5135 } 5136 5137 if (basic.val & (1ul << 48)) 5138 addr_len = 32; 5139 5140 test_vmcs_addr_values("VM-exit-MSR-store address", 5141 EXIT_MSR_ST_ADDR, 16, false, false, 5142 4, addr_len - 1); 5143 5144 /* 5145 * Check last byte of VM-exit MSR-store address 5146 */ 5147 exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf); 5148 5149 for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len); 5150 i < 64; i++) { 5151 tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) | 5152 1ul << i; 5153 vmcs_write(EXIT_MSR_ST_ADDR, 5154 tmp - (exit_msr_st_cnt * 16 - 1)); 5155 test_vmx_invalid_controls(); 5156 } 5157 5158 vmcs_write(EXI_MSR_ST_CNT, 2); 5159 vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16); 5160 test_vmx_invalid_controls(); 5161 vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32); 5162 test_vmx_valid_controls(); 5163 vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48); 5164 test_vmx_valid_controls(); 5165 } 5166 5167 /* 5168 * Tests for VM-exit controls 5169 */ 5170 static void test_vm_exit_ctls(void) 5171 { 5172 test_exit_msr_store(); 5173 } 5174 5175 /* 5176 * Check that the virtual CPU checks all of the VMX controls as 5177 * documented in the Intel SDM. 5178 */ 5179 static void vmx_controls_test(void) 5180 { 5181 /* 5182 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will 5183 * fail due to invalid guest state, should we make it that 5184 * far. 5185 */ 5186 vmcs_write(GUEST_RFLAGS, 0); 5187 5188 test_vm_execution_ctls(); 5189 test_vm_exit_ctls(); 5190 test_vm_entry_ctls(); 5191 } 5192 5193 struct apic_reg_virt_config { 5194 bool apic_register_virtualization; 5195 bool use_tpr_shadow; 5196 bool virtualize_apic_accesses; 5197 bool virtualize_x2apic_mode; 5198 bool activate_secondary_controls; 5199 }; 5200 5201 struct apic_reg_test { 5202 const char *name; 5203 struct apic_reg_virt_config apic_reg_virt_config; 5204 }; 5205 5206 struct apic_reg_virt_expectation { 5207 enum Reason rd_exit_reason; 5208 enum Reason wr_exit_reason; 5209 u32 val; 5210 u32 (*virt_fn)(u32); 5211 5212 /* 5213 * If false, accessing the APIC access address from L2 is treated as a 5214 * normal memory operation, rather than triggering virtualization. 5215 */ 5216 bool virtualize_apic_accesses; 5217 }; 5218 5219 static u32 apic_virt_identity(u32 val) 5220 { 5221 return val; 5222 } 5223 5224 static u32 apic_virt_nibble1(u32 val) 5225 { 5226 return val & 0xf0; 5227 } 5228 5229 static u32 apic_virt_byte3(u32 val) 5230 { 5231 return val & (0xff << 24); 5232 } 5233 5234 static bool apic_reg_virt_exit_expectation( 5235 u32 reg, struct apic_reg_virt_config *config, 5236 struct apic_reg_virt_expectation *expectation) 5237 { 5238 /* Good configs, where some L2 APIC accesses are virtualized. */ 5239 bool virtualize_apic_accesses_only = 5240 config->virtualize_apic_accesses && 5241 !config->use_tpr_shadow && 5242 !config->apic_register_virtualization && 5243 !config->virtualize_x2apic_mode && 5244 config->activate_secondary_controls; 5245 bool virtualize_apic_accesses_and_use_tpr_shadow = 5246 config->virtualize_apic_accesses && 5247 config->use_tpr_shadow && 5248 !config->apic_register_virtualization && 5249 !config->virtualize_x2apic_mode && 5250 config->activate_secondary_controls; 5251 bool apic_register_virtualization = 5252 config->virtualize_apic_accesses && 5253 config->use_tpr_shadow && 5254 config->apic_register_virtualization && 5255 !config->virtualize_x2apic_mode && 5256 config->activate_secondary_controls; 5257 5258 expectation->val = MAGIC_VAL_1; 5259 expectation->virt_fn = apic_virt_identity; 5260 expectation->virtualize_apic_accesses = 5261 config->virtualize_apic_accesses && 5262 config->activate_secondary_controls; 5263 if (virtualize_apic_accesses_only) { 5264 expectation->rd_exit_reason = VMX_APIC_ACCESS; 5265 expectation->wr_exit_reason = VMX_APIC_ACCESS; 5266 } else if (virtualize_apic_accesses_and_use_tpr_shadow) { 5267 switch (reg) { 5268 case APIC_TASKPRI: 5269 expectation->rd_exit_reason = VMX_VMCALL; 5270 expectation->wr_exit_reason = VMX_VMCALL; 5271 expectation->virt_fn = apic_virt_nibble1; 5272 break; 5273 default: 5274 expectation->rd_exit_reason = VMX_APIC_ACCESS; 5275 expectation->wr_exit_reason = VMX_APIC_ACCESS; 5276 } 5277 } else if (apic_register_virtualization) { 5278 expectation->rd_exit_reason = VMX_VMCALL; 5279 5280 switch (reg) { 5281 case APIC_ID: 5282 case APIC_EOI: 5283 case APIC_LDR: 5284 case APIC_DFR: 5285 case APIC_SPIV: 5286 case APIC_ESR: 5287 case APIC_ICR: 5288 case APIC_LVTT: 5289 case APIC_LVTTHMR: 5290 case APIC_LVTPC: 5291 case APIC_LVT0: 5292 case APIC_LVT1: 5293 case APIC_LVTERR: 5294 case APIC_TMICT: 5295 case APIC_TDCR: 5296 expectation->wr_exit_reason = VMX_APIC_WRITE; 5297 break; 5298 case APIC_LVR: 5299 case APIC_ISR ... APIC_ISR + 0x70: 5300 case APIC_TMR ... APIC_TMR + 0x70: 5301 case APIC_IRR ... APIC_IRR + 0x70: 5302 expectation->wr_exit_reason = VMX_APIC_ACCESS; 5303 break; 5304 case APIC_TASKPRI: 5305 expectation->wr_exit_reason = VMX_VMCALL; 5306 expectation->virt_fn = apic_virt_nibble1; 5307 break; 5308 case APIC_ICR2: 5309 expectation->wr_exit_reason = VMX_VMCALL; 5310 expectation->virt_fn = apic_virt_byte3; 5311 break; 5312 default: 5313 expectation->rd_exit_reason = VMX_APIC_ACCESS; 5314 expectation->wr_exit_reason = VMX_APIC_ACCESS; 5315 } 5316 } else if (!expectation->virtualize_apic_accesses) { 5317 /* 5318 * No APIC registers are directly virtualized. This includes 5319 * VTPR, which can be virtualized through MOV to/from CR8 via 5320 * the use TPR shadow control, but not through directly 5321 * accessing VTPR. 5322 */ 5323 expectation->rd_exit_reason = VMX_VMCALL; 5324 expectation->wr_exit_reason = VMX_VMCALL; 5325 } else { 5326 printf("Cannot parse APIC register virtualization config:\n" 5327 "\tvirtualize_apic_accesses: %d\n" 5328 "\tuse_tpr_shadow: %d\n" 5329 "\tapic_register_virtualization: %d\n" 5330 "\tvirtualize_x2apic_mode: %d\n" 5331 "\tactivate_secondary_controls: %d\n", 5332 config->virtualize_apic_accesses, 5333 config->use_tpr_shadow, 5334 config->apic_register_virtualization, 5335 config->virtualize_x2apic_mode, 5336 config->activate_secondary_controls); 5337 5338 return false; 5339 } 5340 5341 return true; 5342 } 5343 5344 struct apic_reg_test apic_reg_tests[] = { 5345 /* Good configs, where some L2 APIC accesses are virtualized. */ 5346 { 5347 .name = "Virtualize APIC accesses", 5348 .apic_reg_virt_config = { 5349 .virtualize_apic_accesses = true, 5350 .use_tpr_shadow = false, 5351 .apic_register_virtualization = false, 5352 .virtualize_x2apic_mode = false, 5353 .activate_secondary_controls = true, 5354 }, 5355 }, 5356 { 5357 .name = "Virtualize APIC accesses + Use TPR shadow", 5358 .apic_reg_virt_config = { 5359 .virtualize_apic_accesses = true, 5360 .use_tpr_shadow = true, 5361 .apic_register_virtualization = false, 5362 .virtualize_x2apic_mode = false, 5363 .activate_secondary_controls = true, 5364 }, 5365 }, 5366 { 5367 .name = "APIC-register virtualization", 5368 .apic_reg_virt_config = { 5369 .virtualize_apic_accesses = true, 5370 .use_tpr_shadow = true, 5371 .apic_register_virtualization = true, 5372 .virtualize_x2apic_mode = false, 5373 .activate_secondary_controls = true, 5374 }, 5375 }, 5376 5377 /* 5378 * Test that the secondary processor-based VM-execution controls are 5379 * correctly ignored when "activate secondary controls" is disabled. 5380 */ 5381 { 5382 .name = "Activate secondary controls off", 5383 .apic_reg_virt_config = { 5384 .virtualize_apic_accesses = true, 5385 .use_tpr_shadow = false, 5386 .apic_register_virtualization = true, 5387 .virtualize_x2apic_mode = true, 5388 .activate_secondary_controls = false, 5389 }, 5390 }, 5391 { 5392 .name = "Activate secondary controls off + Use TPR shadow", 5393 .apic_reg_virt_config = { 5394 .virtualize_apic_accesses = true, 5395 .use_tpr_shadow = true, 5396 .apic_register_virtualization = true, 5397 .virtualize_x2apic_mode = true, 5398 .activate_secondary_controls = false, 5399 }, 5400 }, 5401 5402 /* 5403 * Test that the APIC access address is treated like an arbitrary memory 5404 * address when "virtualize APIC accesses" is disabled. 5405 */ 5406 { 5407 .name = "Virtualize APIC accesses off + Use TPR shadow", 5408 .apic_reg_virt_config = { 5409 .virtualize_apic_accesses = false, 5410 .use_tpr_shadow = true, 5411 .apic_register_virtualization = true, 5412 .virtualize_x2apic_mode = true, 5413 .activate_secondary_controls = true, 5414 }, 5415 }, 5416 5417 /* 5418 * Test that VM entry fails due to invalid controls when 5419 * "APIC-register virtualization" is enabled while "use TPR shadow" is 5420 * disabled. 5421 */ 5422 { 5423 .name = "APIC-register virtualization + Use TPR shadow off", 5424 .apic_reg_virt_config = { 5425 .virtualize_apic_accesses = true, 5426 .use_tpr_shadow = false, 5427 .apic_register_virtualization = true, 5428 .virtualize_x2apic_mode = false, 5429 .activate_secondary_controls = true, 5430 }, 5431 }, 5432 5433 /* 5434 * Test that VM entry fails due to invalid controls when 5435 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is 5436 * disabled. 5437 */ 5438 { 5439 .name = "Virtualize x2APIC mode + Use TPR shadow off", 5440 .apic_reg_virt_config = { 5441 .virtualize_apic_accesses = false, 5442 .use_tpr_shadow = false, 5443 .apic_register_virtualization = false, 5444 .virtualize_x2apic_mode = true, 5445 .activate_secondary_controls = true, 5446 }, 5447 }, 5448 { 5449 .name = "Virtualize x2APIC mode + Use TPR shadow off v2", 5450 .apic_reg_virt_config = { 5451 .virtualize_apic_accesses = false, 5452 .use_tpr_shadow = false, 5453 .apic_register_virtualization = true, 5454 .virtualize_x2apic_mode = true, 5455 .activate_secondary_controls = true, 5456 }, 5457 }, 5458 5459 /* 5460 * Test that VM entry fails due to invalid controls when 5461 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses" 5462 * is enabled. 5463 */ 5464 { 5465 .name = "Virtualize x2APIC mode + Virtualize APIC accesses", 5466 .apic_reg_virt_config = { 5467 .virtualize_apic_accesses = true, 5468 .use_tpr_shadow = true, 5469 .apic_register_virtualization = false, 5470 .virtualize_x2apic_mode = true, 5471 .activate_secondary_controls = true, 5472 }, 5473 }, 5474 { 5475 .name = "Virtualize x2APIC mode + Virtualize APIC accesses v2", 5476 .apic_reg_virt_config = { 5477 .virtualize_apic_accesses = true, 5478 .use_tpr_shadow = true, 5479 .apic_register_virtualization = true, 5480 .virtualize_x2apic_mode = true, 5481 .activate_secondary_controls = true, 5482 }, 5483 }, 5484 }; 5485 5486 enum Apic_op { 5487 APIC_OP_XAPIC_RD, 5488 APIC_OP_XAPIC_WR, 5489 TERMINATE, 5490 }; 5491 5492 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg) 5493 { 5494 return *(volatile u32 *)((uintptr_t)apic_access_address + reg); 5495 } 5496 5497 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val) 5498 { 5499 *(volatile u32 *)((uintptr_t)apic_access_address + reg) = val; 5500 } 5501 5502 struct apic_reg_virt_guest_args { 5503 enum Apic_op op; 5504 u32 *apic_access_address; 5505 u32 reg; 5506 u32 val; 5507 bool check_rd; 5508 u32 (*virt_fn)(u32); 5509 } apic_reg_virt_guest_args; 5510 5511 static void apic_reg_virt_guest(void) 5512 { 5513 volatile struct apic_reg_virt_guest_args *args = 5514 &apic_reg_virt_guest_args; 5515 5516 for (;;) { 5517 enum Apic_op op = args->op; 5518 u32 *apic_access_address = args->apic_access_address; 5519 u32 reg = args->reg; 5520 u32 val = args->val; 5521 bool check_rd = args->check_rd; 5522 u32 (*virt_fn)(u32) = args->virt_fn; 5523 5524 if (op == TERMINATE) 5525 break; 5526 5527 if (op == APIC_OP_XAPIC_RD) { 5528 u32 ret = vmx_xapic_read(apic_access_address, reg); 5529 5530 if (check_rd) { 5531 u32 want = virt_fn(val); 5532 u32 got = virt_fn(ret); 5533 5534 report(got == want, 5535 "read 0x%x, expected 0x%x.", got, want); 5536 } 5537 } else if (op == APIC_OP_XAPIC_WR) { 5538 vmx_xapic_write(apic_access_address, reg, val); 5539 } 5540 5541 /* 5542 * The L1 should always execute a vmcall after it's done testing 5543 * an individual APIC operation. This helps to validate that the 5544 * L1 and L2 are in sync with each other, as expected. 5545 */ 5546 vmcall(); 5547 } 5548 } 5549 5550 static void test_xapic_rd( 5551 u32 reg, struct apic_reg_virt_expectation *expectation, 5552 u32 *apic_access_address, u32 *virtual_apic_page) 5553 { 5554 u32 val = expectation->val; 5555 u32 exit_reason_want = expectation->rd_exit_reason; 5556 struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args; 5557 5558 report_prefix_pushf("xapic - reading 0x%03x", reg); 5559 5560 /* Configure guest to do an xapic read */ 5561 args->op = APIC_OP_XAPIC_RD; 5562 args->apic_access_address = apic_access_address; 5563 args->reg = reg; 5564 args->val = val; 5565 args->check_rd = exit_reason_want == VMX_VMCALL; 5566 args->virt_fn = expectation->virt_fn; 5567 5568 /* Setup virtual APIC page */ 5569 if (!expectation->virtualize_apic_accesses) { 5570 apic_access_address[apic_reg_index(reg)] = val; 5571 virtual_apic_page[apic_reg_index(reg)] = 0; 5572 } else if (exit_reason_want == VMX_VMCALL) { 5573 apic_access_address[apic_reg_index(reg)] = 0; 5574 virtual_apic_page[apic_reg_index(reg)] = val; 5575 } 5576 5577 /* Enter guest */ 5578 enter_guest(); 5579 5580 /* 5581 * Validate the behavior and 5582 * pass a magic value back to the guest. 5583 */ 5584 if (exit_reason_want == VMX_APIC_ACCESS) { 5585 u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff; 5586 5587 assert_exit_reason(exit_reason_want); 5588 report(apic_page_offset == reg, 5589 "got APIC access exit @ page offset 0x%03x, want 0x%03x", 5590 apic_page_offset, reg); 5591 skip_exit_insn(); 5592 5593 /* Reenter guest so it can consume/check rcx and exit again. */ 5594 enter_guest(); 5595 } else if (exit_reason_want != VMX_VMCALL) { 5596 report(false, "Oops, bad exit expectation: %u.", 5597 exit_reason_want); 5598 } 5599 5600 skip_exit_vmcall(); 5601 report_prefix_pop(); 5602 } 5603 5604 static void test_xapic_wr( 5605 u32 reg, struct apic_reg_virt_expectation *expectation, 5606 u32 *apic_access_address, u32 *virtual_apic_page) 5607 { 5608 u32 val = expectation->val; 5609 u32 exit_reason_want = expectation->wr_exit_reason; 5610 struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args; 5611 bool virtualized = 5612 expectation->virtualize_apic_accesses && 5613 (exit_reason_want == VMX_APIC_WRITE || 5614 exit_reason_want == VMX_VMCALL); 5615 bool checked = false; 5616 5617 report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg); 5618 5619 /* Configure guest to do an xapic read */ 5620 args->op = APIC_OP_XAPIC_WR; 5621 args->apic_access_address = apic_access_address; 5622 args->reg = reg; 5623 args->val = val; 5624 5625 /* Setup virtual APIC page */ 5626 if (virtualized || !expectation->virtualize_apic_accesses) { 5627 apic_access_address[apic_reg_index(reg)] = 0; 5628 virtual_apic_page[apic_reg_index(reg)] = 0; 5629 } 5630 5631 /* Enter guest */ 5632 enter_guest(); 5633 5634 /* 5635 * Validate the behavior and 5636 * pass a magic value back to the guest. 5637 */ 5638 if (exit_reason_want == VMX_APIC_ACCESS) { 5639 u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff; 5640 5641 assert_exit_reason(exit_reason_want); 5642 report(apic_page_offset == reg, 5643 "got APIC access exit @ page offset 0x%03x, want 0x%03x", 5644 apic_page_offset, reg); 5645 skip_exit_insn(); 5646 5647 /* Reenter guest so it can consume/check rcx and exit again. */ 5648 enter_guest(); 5649 } else if (exit_reason_want == VMX_APIC_WRITE) { 5650 assert_exit_reason(exit_reason_want); 5651 report(virtual_apic_page[apic_reg_index(reg)] == val, 5652 "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x", 5653 apic_reg_index(reg), 5654 virtual_apic_page[apic_reg_index(reg)], val); 5655 checked = true; 5656 5657 /* Reenter guest so it can consume/check rcx and exit again. */ 5658 enter_guest(); 5659 } else if (exit_reason_want != VMX_VMCALL) { 5660 report(false, "Oops, bad exit expectation: %u.", 5661 exit_reason_want); 5662 } 5663 5664 assert_exit_reason(VMX_VMCALL); 5665 if (virtualized && !checked) { 5666 u32 want = expectation->virt_fn(val); 5667 u32 got = virtual_apic_page[apic_reg_index(reg)]; 5668 got = expectation->virt_fn(got); 5669 5670 report(got == want, "exitless write; val is 0x%x, want 0x%x", 5671 got, want); 5672 } else if (!expectation->virtualize_apic_accesses && !checked) { 5673 u32 got = apic_access_address[apic_reg_index(reg)]; 5674 5675 report(got == val, 5676 "non-virtualized write; val is 0x%x, want 0x%x", got, 5677 val); 5678 } else if (!expectation->virtualize_apic_accesses && checked) { 5679 report(false, 5680 "Non-virtualized write was prematurely checked!"); 5681 } 5682 5683 skip_exit_vmcall(); 5684 report_prefix_pop(); 5685 } 5686 5687 enum Config_type { 5688 CONFIG_TYPE_GOOD, 5689 CONFIG_TYPE_UNSUPPORTED, 5690 CONFIG_TYPE_VMENTRY_FAILS_EARLY, 5691 }; 5692 5693 static enum Config_type configure_apic_reg_virt_test( 5694 struct apic_reg_virt_config *apic_reg_virt_config) 5695 { 5696 u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0); 5697 u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1); 5698 /* Configs where L2 entry fails early, due to invalid controls. */ 5699 bool use_tpr_shadow_incorrectly_off = 5700 !apic_reg_virt_config->use_tpr_shadow && 5701 (apic_reg_virt_config->apic_register_virtualization || 5702 apic_reg_virt_config->virtualize_x2apic_mode) && 5703 apic_reg_virt_config->activate_secondary_controls; 5704 bool virtualize_apic_accesses_incorrectly_on = 5705 apic_reg_virt_config->virtualize_apic_accesses && 5706 apic_reg_virt_config->virtualize_x2apic_mode && 5707 apic_reg_virt_config->activate_secondary_controls; 5708 bool vmentry_fails_early = 5709 use_tpr_shadow_incorrectly_off || 5710 virtualize_apic_accesses_incorrectly_on; 5711 5712 if (apic_reg_virt_config->activate_secondary_controls) { 5713 if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) { 5714 printf("VM-execution control \"activate secondary controls\" NOT supported.\n"); 5715 return CONFIG_TYPE_UNSUPPORTED; 5716 } 5717 cpu_exec_ctrl0 |= CPU_SECONDARY; 5718 } else { 5719 cpu_exec_ctrl0 &= ~CPU_SECONDARY; 5720 } 5721 5722 if (apic_reg_virt_config->virtualize_apic_accesses) { 5723 if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) { 5724 printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n"); 5725 return CONFIG_TYPE_UNSUPPORTED; 5726 } 5727 cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES; 5728 } else { 5729 cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES; 5730 } 5731 5732 if (apic_reg_virt_config->use_tpr_shadow) { 5733 if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) { 5734 printf("VM-execution control \"use TPR shadow\" NOT supported.\n"); 5735 return CONFIG_TYPE_UNSUPPORTED; 5736 } 5737 cpu_exec_ctrl0 |= CPU_TPR_SHADOW; 5738 } else { 5739 cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW; 5740 } 5741 5742 if (apic_reg_virt_config->apic_register_virtualization) { 5743 if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) { 5744 printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n"); 5745 return CONFIG_TYPE_UNSUPPORTED; 5746 } 5747 cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT; 5748 } else { 5749 cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT; 5750 } 5751 5752 if (apic_reg_virt_config->virtualize_x2apic_mode) { 5753 if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) { 5754 printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n"); 5755 return CONFIG_TYPE_UNSUPPORTED; 5756 } 5757 cpu_exec_ctrl1 |= CPU_VIRT_X2APIC; 5758 } else { 5759 cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC; 5760 } 5761 5762 vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0); 5763 vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1); 5764 5765 if (vmentry_fails_early) 5766 return CONFIG_TYPE_VMENTRY_FAILS_EARLY; 5767 5768 return CONFIG_TYPE_GOOD; 5769 } 5770 5771 static bool cpu_has_apicv(void) 5772 { 5773 return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) && 5774 (ctrl_cpu_rev[1].clr & CPU_VINTD) && 5775 (ctrl_pin_rev.clr & PIN_POST_INTR)); 5776 } 5777 5778 /* Validates APIC register access across valid virtualization configurations. */ 5779 static void apic_reg_virt_test(void) 5780 { 5781 u32 *apic_access_address; 5782 u32 *virtual_apic_page; 5783 u64 control; 5784 u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0); 5785 u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1); 5786 int i; 5787 struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args; 5788 5789 if (!cpu_has_apicv()) { 5790 report_skip(__func__); 5791 return; 5792 } 5793 5794 control = cpu_exec_ctrl1; 5795 control &= ~CPU_VINTD; 5796 vmcs_write(CPU_EXEC_CTRL1, control); 5797 5798 test_set_guest(apic_reg_virt_guest); 5799 5800 /* 5801 * From the SDM: The 1-setting of the "virtualize APIC accesses" 5802 * VM-execution is guaranteed to apply only if translations to the 5803 * APIC-access address use a 4-KByte page. 5804 */ 5805 apic_access_address = alloc_page(); 5806 force_4k_page(apic_access_address); 5807 vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address)); 5808 5809 virtual_apic_page = alloc_page(); 5810 vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page)); 5811 5812 for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) { 5813 struct apic_reg_test *apic_reg_test = &apic_reg_tests[i]; 5814 struct apic_reg_virt_config *apic_reg_virt_config = 5815 &apic_reg_test->apic_reg_virt_config; 5816 enum Config_type config_type; 5817 u32 reg; 5818 5819 printf("--- %s test ---\n", apic_reg_test->name); 5820 config_type = 5821 configure_apic_reg_virt_test(apic_reg_virt_config); 5822 if (config_type == CONFIG_TYPE_UNSUPPORTED) { 5823 printf("Skip because of missing features.\n"); 5824 continue; 5825 } 5826 5827 if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) { 5828 enter_guest_with_bad_controls(); 5829 continue; 5830 } 5831 5832 for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) { 5833 struct apic_reg_virt_expectation expectation = {}; 5834 bool ok; 5835 5836 ok = apic_reg_virt_exit_expectation( 5837 reg, apic_reg_virt_config, &expectation); 5838 if (!ok) { 5839 report(false, "Malformed test."); 5840 break; 5841 } 5842 5843 test_xapic_rd(reg, &expectation, apic_access_address, 5844 virtual_apic_page); 5845 test_xapic_wr(reg, &expectation, apic_access_address, 5846 virtual_apic_page); 5847 } 5848 } 5849 5850 /* Terminate the guest */ 5851 vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0); 5852 vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1); 5853 args->op = TERMINATE; 5854 enter_guest(); 5855 assert_exit_reason(VMX_VMCALL); 5856 } 5857 5858 struct virt_x2apic_mode_config { 5859 struct apic_reg_virt_config apic_reg_virt_config; 5860 bool virtual_interrupt_delivery; 5861 bool use_msr_bitmaps; 5862 bool disable_x2apic_msr_intercepts; 5863 bool disable_x2apic; 5864 }; 5865 5866 struct virt_x2apic_mode_test_case { 5867 const char *name; 5868 struct virt_x2apic_mode_config virt_x2apic_mode_config; 5869 }; 5870 5871 enum Virt_x2apic_mode_behavior_type { 5872 X2APIC_ACCESS_VIRTUALIZED, 5873 X2APIC_ACCESS_PASSED_THROUGH, 5874 X2APIC_ACCESS_TRIGGERS_GP, 5875 }; 5876 5877 struct virt_x2apic_mode_expectation { 5878 enum Reason rd_exit_reason; 5879 enum Reason wr_exit_reason; 5880 5881 /* 5882 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of 5883 * the x2APIC registers are 32 bits. Notice: 5884 * 1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers. 5885 * 2. vmx_x2apic_write() expects the val arg to be well-formed. 5886 */ 5887 u64 rd_val; 5888 u64 wr_val; 5889 5890 /* 5891 * Compares input to virtualized output; 5892 * 1st arg is pointer to return expected virtualization output. 5893 */ 5894 u64 (*virt_fn)(u64); 5895 5896 enum Virt_x2apic_mode_behavior_type rd_behavior; 5897 enum Virt_x2apic_mode_behavior_type wr_behavior; 5898 bool wr_only; 5899 }; 5900 5901 static u64 virt_x2apic_mode_identity(u64 val) 5902 { 5903 return val; 5904 } 5905 5906 static u64 virt_x2apic_mode_nibble1(u64 val) 5907 { 5908 return val & 0xf0; 5909 } 5910 5911 static void virt_x2apic_mode_rd_expectation( 5912 u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic, 5913 bool apic_register_virtualization, bool virtual_interrupt_delivery, 5914 struct virt_x2apic_mode_expectation *expectation) 5915 { 5916 bool readable = 5917 !x2apic_reg_reserved(reg) && 5918 reg != APIC_EOI; 5919 5920 expectation->rd_exit_reason = VMX_VMCALL; 5921 expectation->virt_fn = virt_x2apic_mode_identity; 5922 if (virt_x2apic_mode_on && apic_register_virtualization) { 5923 expectation->rd_val = MAGIC_VAL_1; 5924 if (reg == APIC_PROCPRI && virtual_interrupt_delivery) 5925 expectation->virt_fn = virt_x2apic_mode_nibble1; 5926 else if (reg == APIC_TASKPRI) 5927 expectation->virt_fn = virt_x2apic_mode_nibble1; 5928 expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED; 5929 } else if (virt_x2apic_mode_on && !apic_register_virtualization && 5930 reg == APIC_TASKPRI) { 5931 expectation->rd_val = MAGIC_VAL_1; 5932 expectation->virt_fn = virt_x2apic_mode_nibble1; 5933 expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED; 5934 } else if (!disable_x2apic && readable) { 5935 expectation->rd_val = apic_read(reg); 5936 expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH; 5937 } else { 5938 expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP; 5939 } 5940 } 5941 5942 /* 5943 * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register. 5944 * 5945 * For writable registers, get_x2apic_wr_val() deposits the write value into the 5946 * val pointer arg and returns true. For non-writable registers, val is not 5947 * modified and get_x2apic_wr_val() returns false. 5948 */ 5949 static bool get_x2apic_wr_val(u32 reg, u64 *val) 5950 { 5951 switch (reg) { 5952 case APIC_TASKPRI: 5953 /* Bits 31:8 are reserved. */ 5954 *val &= 0xff; 5955 break; 5956 case APIC_EOI: 5957 case APIC_ESR: 5958 case APIC_TMICT: 5959 /* 5960 * EOI, ESR: WRMSR of a non-zero value causes #GP(0). 5961 * TMICT: A write of 0 to the initial-count register effectively 5962 * stops the local APIC timer, in both one-shot and 5963 * periodic mode. 5964 */ 5965 *val = 0; 5966 break; 5967 case APIC_SPIV: 5968 case APIC_LVTT: 5969 case APIC_LVTTHMR: 5970 case APIC_LVTPC: 5971 case APIC_LVT0: 5972 case APIC_LVT1: 5973 case APIC_LVTERR: 5974 case APIC_TDCR: 5975 /* 5976 * To avoid writing a 1 to a reserved bit or causing some other 5977 * unintended side effect, read the current value and use it as 5978 * the write value. 5979 */ 5980 *val = apic_read(reg); 5981 break; 5982 case APIC_CMCI: 5983 if (!apic_lvt_entry_supported(6)) 5984 return false; 5985 *val = apic_read(reg); 5986 break; 5987 case APIC_ICR: 5988 *val = 0x40000 | 0xf1; 5989 break; 5990 case APIC_SELF_IPI: 5991 /* 5992 * With special processing (i.e., virtualize x2APIC mode + 5993 * virtual interrupt delivery), writing zero causes an 5994 * APIC-write VM exit. We plan to add a test for enabling 5995 * "virtual-interrupt delivery" in VMCS12, and that's where we 5996 * will test a self IPI with special processing. 5997 */ 5998 *val = 0x0; 5999 break; 6000 default: 6001 return false; 6002 } 6003 6004 return true; 6005 } 6006 6007 static bool special_processing_applies(u32 reg, u64 *val, 6008 bool virt_int_delivery) 6009 { 6010 bool special_processing = 6011 (reg == APIC_TASKPRI) || 6012 (virt_int_delivery && 6013 (reg == APIC_EOI || reg == APIC_SELF_IPI)); 6014 6015 if (special_processing) { 6016 TEST_ASSERT(get_x2apic_wr_val(reg, val)); 6017 return true; 6018 } 6019 6020 return false; 6021 } 6022 6023 static void virt_x2apic_mode_wr_expectation( 6024 u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic, 6025 bool virt_int_delivery, 6026 struct virt_x2apic_mode_expectation *expectation) 6027 { 6028 expectation->wr_exit_reason = VMX_VMCALL; 6029 expectation->wr_val = MAGIC_VAL_1; 6030 expectation->wr_only = false; 6031 6032 if (virt_x2apic_mode_on && 6033 special_processing_applies(reg, &expectation->wr_val, 6034 virt_int_delivery)) { 6035 expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED; 6036 if (reg == APIC_SELF_IPI) 6037 expectation->wr_exit_reason = VMX_APIC_WRITE; 6038 } else if (!disable_x2apic && 6039 get_x2apic_wr_val(reg, &expectation->wr_val)) { 6040 expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH; 6041 if (reg == APIC_EOI || reg == APIC_SELF_IPI) 6042 expectation->wr_only = true; 6043 if (reg == APIC_ICR) 6044 expectation->wr_exit_reason = VMX_EXTINT; 6045 } else { 6046 expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP; 6047 /* 6048 * Writing 1 to a reserved bit triggers a #GP. 6049 * Thus, set the write value to 0, which seems 6050 * the most likely to detect a missed #GP. 6051 */ 6052 expectation->wr_val = 0; 6053 } 6054 } 6055 6056 static void virt_x2apic_mode_exit_expectation( 6057 u32 reg, struct virt_x2apic_mode_config *config, 6058 struct virt_x2apic_mode_expectation *expectation) 6059 { 6060 struct apic_reg_virt_config *base_config = 6061 &config->apic_reg_virt_config; 6062 bool virt_x2apic_mode_on = 6063 base_config->virtualize_x2apic_mode && 6064 config->use_msr_bitmaps && 6065 config->disable_x2apic_msr_intercepts && 6066 base_config->activate_secondary_controls; 6067 6068 virt_x2apic_mode_wr_expectation( 6069 reg, virt_x2apic_mode_on, config->disable_x2apic, 6070 config->virtual_interrupt_delivery, expectation); 6071 virt_x2apic_mode_rd_expectation( 6072 reg, virt_x2apic_mode_on, config->disable_x2apic, 6073 base_config->apic_register_virtualization, 6074 config->virtual_interrupt_delivery, expectation); 6075 } 6076 6077 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = { 6078 /* 6079 * Baseline "virtualize x2APIC mode" configuration: 6080 * - virtualize x2APIC mode 6081 * - virtual-interrupt delivery 6082 * - APIC-register virtualization 6083 * - x2APIC MSR intercepts disabled 6084 * 6085 * Reads come from virtual APIC page, special processing applies to 6086 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1 6087 * APIC. 6088 */ 6089 { 6090 .name = "Baseline", 6091 .virt_x2apic_mode_config = { 6092 .virtual_interrupt_delivery = true, 6093 .use_msr_bitmaps = true, 6094 .disable_x2apic_msr_intercepts = true, 6095 .disable_x2apic = false, 6096 .apic_reg_virt_config = { 6097 .apic_register_virtualization = true, 6098 .use_tpr_shadow = true, 6099 .virtualize_apic_accesses = false, 6100 .virtualize_x2apic_mode = true, 6101 .activate_secondary_controls = true, 6102 }, 6103 }, 6104 }, 6105 { 6106 .name = "Baseline w/ x2apic disabled", 6107 .virt_x2apic_mode_config = { 6108 .virtual_interrupt_delivery = true, 6109 .use_msr_bitmaps = true, 6110 .disable_x2apic_msr_intercepts = true, 6111 .disable_x2apic = true, 6112 .apic_reg_virt_config = { 6113 .apic_register_virtualization = true, 6114 .use_tpr_shadow = true, 6115 .virtualize_apic_accesses = false, 6116 .virtualize_x2apic_mode = true, 6117 .activate_secondary_controls = true, 6118 }, 6119 }, 6120 }, 6121 6122 /* 6123 * Baseline, minus virtual-interrupt delivery. Reads come from virtual 6124 * APIC page, special processing applies to VTPR, and all other writes 6125 * pass through to L1 APIC. 6126 */ 6127 { 6128 .name = "Baseline - virtual interrupt delivery", 6129 .virt_x2apic_mode_config = { 6130 .virtual_interrupt_delivery = false, 6131 .use_msr_bitmaps = true, 6132 .disable_x2apic_msr_intercepts = true, 6133 .disable_x2apic = false, 6134 .apic_reg_virt_config = { 6135 .apic_register_virtualization = true, 6136 .use_tpr_shadow = true, 6137 .virtualize_apic_accesses = false, 6138 .virtualize_x2apic_mode = true, 6139 .activate_secondary_controls = true, 6140 }, 6141 }, 6142 }, 6143 6144 /* 6145 * Baseline, minus APIC-register virtualization. x2APIC reads pass 6146 * through to L1's APIC, unless reading VTPR 6147 */ 6148 { 6149 .name = "Virtualize x2APIC mode, no APIC reg virt", 6150 .virt_x2apic_mode_config = { 6151 .virtual_interrupt_delivery = true, 6152 .use_msr_bitmaps = true, 6153 .disable_x2apic_msr_intercepts = true, 6154 .disable_x2apic = false, 6155 .apic_reg_virt_config = { 6156 .apic_register_virtualization = false, 6157 .use_tpr_shadow = true, 6158 .virtualize_apic_accesses = false, 6159 .virtualize_x2apic_mode = true, 6160 .activate_secondary_controls = true, 6161 }, 6162 }, 6163 }, 6164 { 6165 .name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off", 6166 .virt_x2apic_mode_config = { 6167 .virtual_interrupt_delivery = true, 6168 .use_msr_bitmaps = true, 6169 .disable_x2apic_msr_intercepts = true, 6170 .disable_x2apic = true, 6171 .apic_reg_virt_config = { 6172 .apic_register_virtualization = false, 6173 .use_tpr_shadow = true, 6174 .virtualize_apic_accesses = false, 6175 .virtualize_x2apic_mode = true, 6176 .activate_secondary_controls = true, 6177 }, 6178 }, 6179 }, 6180 6181 /* 6182 * Enable "virtualize x2APIC mode" and "APIC-register virtualization", 6183 * and disable intercepts for the x2APIC MSRs, but fail to enable 6184 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC 6185 * MSRs). 6186 */ 6187 { 6188 .name = "Fail to enable activate secondary controls", 6189 .virt_x2apic_mode_config = { 6190 .virtual_interrupt_delivery = true, 6191 .use_msr_bitmaps = true, 6192 .disable_x2apic_msr_intercepts = true, 6193 .disable_x2apic = false, 6194 .apic_reg_virt_config = { 6195 .apic_register_virtualization = true, 6196 .use_tpr_shadow = true, 6197 .virtualize_apic_accesses = false, 6198 .virtualize_x2apic_mode = true, 6199 .activate_secondary_controls = false, 6200 }, 6201 }, 6202 }, 6203 6204 /* 6205 * Enable "APIC-register virtualization" and enable "activate secondary 6206 * controls" and disable intercepts for the x2APIC MSRs, but do not 6207 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2 6208 * gets access to L1's x2APIC MSRs). 6209 */ 6210 { 6211 .name = "Fail to enable virtualize x2APIC mode", 6212 .virt_x2apic_mode_config = { 6213 .virtual_interrupt_delivery = true, 6214 .use_msr_bitmaps = true, 6215 .disable_x2apic_msr_intercepts = true, 6216 .disable_x2apic = false, 6217 .apic_reg_virt_config = { 6218 .apic_register_virtualization = true, 6219 .use_tpr_shadow = true, 6220 .virtualize_apic_accesses = false, 6221 .virtualize_x2apic_mode = false, 6222 .activate_secondary_controls = true, 6223 }, 6224 }, 6225 }, 6226 6227 /* 6228 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and 6229 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs. 6230 */ 6231 { 6232 .name = "Baseline", 6233 .virt_x2apic_mode_config = { 6234 .virtual_interrupt_delivery = true, 6235 .use_msr_bitmaps = true, 6236 .disable_x2apic_msr_intercepts = true, 6237 .disable_x2apic = false, 6238 .apic_reg_virt_config = { 6239 .apic_register_virtualization = true, 6240 .use_tpr_shadow = true, 6241 .virtualize_apic_accesses = false, 6242 .virtualize_x2apic_mode = false, 6243 .activate_secondary_controls = true, 6244 }, 6245 }, 6246 }, 6247 }; 6248 6249 enum X2apic_op { 6250 X2APIC_OP_RD, 6251 X2APIC_OP_WR, 6252 X2APIC_TERMINATE, 6253 }; 6254 6255 static u64 vmx_x2apic_read(u32 reg) 6256 { 6257 u32 msr_addr = x2apic_msr(reg); 6258 u64 val; 6259 6260 val = rdmsr(msr_addr); 6261 6262 return val; 6263 } 6264 6265 static void vmx_x2apic_write(u32 reg, u64 val) 6266 { 6267 u32 msr_addr = x2apic_msr(reg); 6268 6269 wrmsr(msr_addr, val); 6270 } 6271 6272 struct virt_x2apic_mode_guest_args { 6273 enum X2apic_op op; 6274 u32 reg; 6275 u64 val; 6276 bool should_gp; 6277 u64 (*virt_fn)(u64); 6278 } virt_x2apic_mode_guest_args; 6279 6280 static volatile bool handle_x2apic_gp_ran; 6281 static volatile u32 handle_x2apic_gp_insn_len; 6282 static void handle_x2apic_gp(struct ex_regs *regs) 6283 { 6284 handle_x2apic_gp_ran = true; 6285 regs->rip += handle_x2apic_gp_insn_len; 6286 } 6287 6288 static handler setup_x2apic_gp_handler(void) 6289 { 6290 handler old_handler; 6291 6292 old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp); 6293 /* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */ 6294 handle_x2apic_gp_insn_len = 2; 6295 6296 return old_handler; 6297 } 6298 6299 static void teardown_x2apic_gp_handler(handler old_handler) 6300 { 6301 handle_exception(GP_VECTOR, old_handler); 6302 6303 /* 6304 * Defensively reset instruction length, so that if the handler is 6305 * incorrectly used, it will loop infinitely, rather than run off into 6306 * la la land. 6307 */ 6308 handle_x2apic_gp_insn_len = 0; 6309 handle_x2apic_gp_ran = false; 6310 } 6311 6312 static void virt_x2apic_mode_guest(void) 6313 { 6314 volatile struct virt_x2apic_mode_guest_args *args = 6315 &virt_x2apic_mode_guest_args; 6316 6317 for (;;) { 6318 enum X2apic_op op = args->op; 6319 u32 reg = args->reg; 6320 u64 val = args->val; 6321 bool should_gp = args->should_gp; 6322 u64 (*virt_fn)(u64) = args->virt_fn; 6323 handler old_handler; 6324 6325 if (op == X2APIC_TERMINATE) 6326 break; 6327 6328 if (should_gp) { 6329 TEST_ASSERT(!handle_x2apic_gp_ran); 6330 old_handler = setup_x2apic_gp_handler(); 6331 } 6332 6333 if (op == X2APIC_OP_RD) { 6334 u64 ret = vmx_x2apic_read(reg); 6335 6336 if (!should_gp) { 6337 u64 want = virt_fn(val); 6338 u64 got = virt_fn(ret); 6339 6340 report(got == want, 6341 "APIC read; got 0x%lx, want 0x%lx.", 6342 got, want); 6343 } 6344 } else if (op == X2APIC_OP_WR) { 6345 vmx_x2apic_write(reg, val); 6346 } 6347 6348 if (should_gp) { 6349 report(handle_x2apic_gp_ran, 6350 "x2APIC op triggered GP."); 6351 teardown_x2apic_gp_handler(old_handler); 6352 } 6353 6354 /* 6355 * The L1 should always execute a vmcall after it's done testing 6356 * an individual APIC operation. This helps to validate that the 6357 * L1 and L2 are in sync with each other, as expected. 6358 */ 6359 vmcall(); 6360 } 6361 } 6362 6363 static void test_x2apic_rd( 6364 u32 reg, struct virt_x2apic_mode_expectation *expectation, 6365 u32 *virtual_apic_page) 6366 { 6367 u64 val = expectation->rd_val; 6368 u32 exit_reason_want = expectation->rd_exit_reason; 6369 struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args; 6370 6371 report_prefix_pushf("x2apic - reading 0x%03x", reg); 6372 6373 /* Configure guest to do an x2apic read */ 6374 args->op = X2APIC_OP_RD; 6375 args->reg = reg; 6376 args->val = val; 6377 args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP; 6378 args->virt_fn = expectation->virt_fn; 6379 6380 /* Setup virtual APIC page */ 6381 if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED) 6382 virtual_apic_page[apic_reg_index(reg)] = (u32)val; 6383 6384 /* Enter guest */ 6385 enter_guest(); 6386 6387 if (exit_reason_want != VMX_VMCALL) { 6388 report(false, "Oops, bad exit expectation: %u.", 6389 exit_reason_want); 6390 } 6391 6392 skip_exit_vmcall(); 6393 report_prefix_pop(); 6394 } 6395 6396 static volatile bool handle_x2apic_ipi_ran; 6397 static void handle_x2apic_ipi(isr_regs_t *regs) 6398 { 6399 handle_x2apic_ipi_ran = true; 6400 eoi(); 6401 } 6402 6403 static void test_x2apic_wr( 6404 u32 reg, struct virt_x2apic_mode_expectation *expectation, 6405 u32 *virtual_apic_page) 6406 { 6407 u64 val = expectation->wr_val; 6408 u32 exit_reason_want = expectation->wr_exit_reason; 6409 struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args; 6410 int ipi_vector = 0xf1; 6411 u32 restore_val = 0; 6412 6413 report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg); 6414 6415 /* Configure guest to do an x2apic read */ 6416 args->op = X2APIC_OP_WR; 6417 args->reg = reg; 6418 args->val = val; 6419 args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP; 6420 6421 /* Setup virtual APIC page */ 6422 if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) 6423 virtual_apic_page[apic_reg_index(reg)] = 0; 6424 if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only) 6425 restore_val = apic_read(reg); 6426 6427 /* Setup IPI handler */ 6428 handle_x2apic_ipi_ran = false; 6429 handle_irq(ipi_vector, handle_x2apic_ipi); 6430 6431 /* Enter guest */ 6432 enter_guest(); 6433 6434 /* 6435 * Validate the behavior and 6436 * pass a magic value back to the guest. 6437 */ 6438 if (exit_reason_want == VMX_EXTINT) { 6439 assert_exit_reason(exit_reason_want); 6440 6441 /* Clear the external interrupt. */ 6442 irq_enable(); 6443 asm volatile ("nop"); 6444 irq_disable(); 6445 report(handle_x2apic_ipi_ran, 6446 "Got pending interrupt after IRQ enabled."); 6447 6448 enter_guest(); 6449 } else if (exit_reason_want == VMX_APIC_WRITE) { 6450 assert_exit_reason(exit_reason_want); 6451 report(virtual_apic_page[apic_reg_index(reg)] == val, 6452 "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx", 6453 apic_reg_index(reg), 6454 virtual_apic_page[apic_reg_index(reg)], val); 6455 6456 /* Reenter guest so it can consume/check rcx and exit again. */ 6457 enter_guest(); 6458 } else if (exit_reason_want != VMX_VMCALL) { 6459 report(false, "Oops, bad exit expectation: %u.", 6460 exit_reason_want); 6461 } 6462 6463 assert_exit_reason(VMX_VMCALL); 6464 if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) { 6465 u64 want = val; 6466 u32 got = virtual_apic_page[apic_reg_index(reg)]; 6467 6468 report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got, 6469 want); 6470 } else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) { 6471 if (!expectation->wr_only) { 6472 u32 got = apic_read(reg); 6473 bool ok; 6474 6475 /* 6476 * When L1's TPR is passed through to L2, the lower 6477 * nibble can be lost. For example, if L2 executes 6478 * WRMSR(0x808, 0x78), then, L1 might read 0x70. 6479 * 6480 * Here's how the lower nibble can get lost: 6481 * 1. L2 executes WRMSR(0x808, 0x78). 6482 * 2. L2 exits to L0 with a WRMSR exit. 6483 * 3. L0 emulates WRMSR, by writing L1's TPR. 6484 * 4. L0 re-enters L2. 6485 * 5. L2 exits to L0 (reason doesn't matter). 6486 * 6. L0 reflects L2's exit to L1. 6487 * 7. Before entering L1, L0 exits to user-space 6488 * (e.g., to satisfy TPR access reporting). 6489 * 8. User-space executes KVM_SET_REGS ioctl, which 6490 * clears the lower nibble of L1's TPR. 6491 */ 6492 if (reg == APIC_TASKPRI) { 6493 got = apic_virt_nibble1(got); 6494 val = apic_virt_nibble1(val); 6495 } 6496 6497 ok = got == val; 6498 report(ok, 6499 "non-virtualized write; val is 0x%x, want 0x%lx", 6500 got, val); 6501 apic_write(reg, restore_val); 6502 } else { 6503 report(true, "non-virtualized and write-only OK"); 6504 } 6505 } 6506 skip_exit_insn(); 6507 6508 report_prefix_pop(); 6509 } 6510 6511 static enum Config_type configure_virt_x2apic_mode_test( 6512 struct virt_x2apic_mode_config *virt_x2apic_mode_config, 6513 u8 *msr_bitmap_page) 6514 { 6515 int msr; 6516 u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0); 6517 u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1); 6518 6519 /* x2apic-specific VMCS config */ 6520 if (virt_x2apic_mode_config->use_msr_bitmaps) { 6521 /* virt_x2apic_mode_test() checks for MSR bitmaps support */ 6522 cpu_exec_ctrl0 |= CPU_MSR_BITMAP; 6523 } else { 6524 cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP; 6525 } 6526 6527 if (virt_x2apic_mode_config->virtual_interrupt_delivery) { 6528 if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) { 6529 report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n"); 6530 return CONFIG_TYPE_UNSUPPORTED; 6531 } 6532 cpu_exec_ctrl1 |= CPU_VINTD; 6533 } else { 6534 cpu_exec_ctrl1 &= ~CPU_VINTD; 6535 } 6536 6537 vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0); 6538 vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1); 6539 6540 /* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */ 6541 for (msr = 0x800; msr <= 0x8ff; msr++) { 6542 if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) { 6543 clear_bit(msr, msr_bitmap_page + 0x000); 6544 clear_bit(msr, msr_bitmap_page + 0x800); 6545 } else { 6546 set_bit(msr, msr_bitmap_page + 0x000); 6547 set_bit(msr, msr_bitmap_page + 0x800); 6548 } 6549 } 6550 6551 /* x2APIC mode can impact virtualization */ 6552 reset_apic(); 6553 if (!virt_x2apic_mode_config->disable_x2apic) 6554 enable_x2apic(); 6555 6556 return configure_apic_reg_virt_test( 6557 &virt_x2apic_mode_config->apic_reg_virt_config); 6558 } 6559 6560 static void virt_x2apic_mode_test(void) 6561 { 6562 u32 *virtual_apic_page; 6563 u8 *msr_bitmap_page; 6564 u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0); 6565 u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1); 6566 int i; 6567 struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args; 6568 6569 if (!cpu_has_apicv()) { 6570 report_skip(__func__); 6571 return; 6572 } 6573 6574 /* 6575 * This is to exercise an issue in KVM's logic to merge L0's and L1's 6576 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by 6577 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would 6578 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the 6579 * "Virtualize x2APIC mod" control set, even when it didn't. 6580 */ 6581 if (has_spec_ctrl()) 6582 wrmsr(MSR_IA32_SPEC_CTRL, 1); 6583 6584 /* 6585 * Check that VMCS12 supports: 6586 * - "Virtual-APIC address", indicated by "use TPR shadow" 6587 * - "MSR-bitmap address", indicated by "use MSR bitmaps" 6588 */ 6589 if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) { 6590 report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n"); 6591 return; 6592 } else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) { 6593 report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n"); 6594 return; 6595 } 6596 6597 test_set_guest(virt_x2apic_mode_guest); 6598 6599 virtual_apic_page = alloc_page(); 6600 vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page)); 6601 6602 msr_bitmap_page = alloc_page(); 6603 memset(msr_bitmap_page, 0xff, PAGE_SIZE); 6604 vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page)); 6605 6606 for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) { 6607 struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case = 6608 &virt_x2apic_mode_tests[i]; 6609 struct virt_x2apic_mode_config *virt_x2apic_mode_config = 6610 &virt_x2apic_mode_test_case->virt_x2apic_mode_config; 6611 enum Config_type config_type; 6612 u32 reg; 6613 6614 printf("--- %s test ---\n", virt_x2apic_mode_test_case->name); 6615 config_type = 6616 configure_virt_x2apic_mode_test(virt_x2apic_mode_config, 6617 msr_bitmap_page); 6618 if (config_type == CONFIG_TYPE_UNSUPPORTED) { 6619 report_skip("Skip because of missing features.\n"); 6620 continue; 6621 } else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) { 6622 enter_guest_with_bad_controls(); 6623 continue; 6624 } 6625 6626 for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) { 6627 struct virt_x2apic_mode_expectation expectation; 6628 6629 virt_x2apic_mode_exit_expectation( 6630 reg, virt_x2apic_mode_config, &expectation); 6631 6632 test_x2apic_rd(reg, &expectation, virtual_apic_page); 6633 test_x2apic_wr(reg, &expectation, virtual_apic_page); 6634 } 6635 } 6636 6637 6638 /* Terminate the guest */ 6639 vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0); 6640 vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1); 6641 args->op = X2APIC_TERMINATE; 6642 enter_guest(); 6643 assert_exit_reason(VMX_VMCALL); 6644 } 6645 6646 /* 6647 * On processors that support Intel 64 architecture, the IA32_SYSENTER_ESP 6648 * field and the IA32_SYSENTER_EIP field must each contain a canonical 6649 * address. 6650 * 6651 * [Intel SDM] 6652 */ 6653 static void test_sysenter_field(u32 field, const char *name) 6654 { 6655 u64 addr_saved = vmcs_read(field); 6656 6657 vmcs_write(field, NONCANONICAL); 6658 report_prefix_pushf("%s non-canonical", name); 6659 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 6660 report_prefix_pop(); 6661 6662 vmcs_write(field, 0xffffffff); 6663 report_prefix_pushf("%s canonical", name); 6664 test_vmx_vmlaunch(0); 6665 report_prefix_pop(); 6666 6667 vmcs_write(field, addr_saved); 6668 } 6669 6670 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1) 6671 { 6672 u64 val; 6673 u64 cr_saved = vmcs_read(cr); 6674 int i; 6675 6676 val = fixed0 & fixed1; 6677 if (cr == HOST_CR4) 6678 vmcs_write(cr, val | X86_CR4_PAE); 6679 else 6680 vmcs_write(cr, val); 6681 report_prefix_pushf("%s %lx", cr_name, val); 6682 if (val == fixed0) 6683 test_vmx_vmlaunch(0); 6684 else 6685 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 6686 report_prefix_pop(); 6687 6688 for (i = 0; i < 64; i++) { 6689 6690 /* Set a bit when the corresponding bit in fixed1 is 0 */ 6691 if ((fixed1 & (1ull << i)) == 0) { 6692 if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP || 6693 (1ull << i) & X86_CR4_SMAP)) 6694 continue; 6695 6696 vmcs_write(cr, cr_saved | (1ull << i)); 6697 report_prefix_pushf("%s %llx", cr_name, 6698 cr_saved | (1ull << i)); 6699 test_vmx_vmlaunch( 6700 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 6701 report_prefix_pop(); 6702 } 6703 6704 /* Unset a bit when the corresponding bit in fixed0 is 1 */ 6705 if (fixed0 & (1ull << i)) { 6706 vmcs_write(cr, cr_saved & ~(1ull << i)); 6707 report_prefix_pushf("%s %llx", cr_name, 6708 cr_saved & ~(1ull << i)); 6709 test_vmx_vmlaunch( 6710 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 6711 report_prefix_pop(); 6712 } 6713 } 6714 6715 vmcs_write(cr, cr_saved); 6716 } 6717 6718 /* 6719 * 1. The CR0 field must not set any bit to a value not supported in VMX 6720 * operation. 6721 * 2. The CR4 field must not set any bit to a value not supported in VMX 6722 * operation. 6723 * 3. On processors that support Intel 64 architecture, the CR3 field must 6724 * be such that bits 63:52 and bits in the range 51:32 beyond the 6725 * processor’s physical-address width must be 0. 6726 * 6727 * [Intel SDM] 6728 */ 6729 static void test_host_ctl_regs(void) 6730 { 6731 u64 fixed0, fixed1, cr3, cr3_saved; 6732 int i; 6733 6734 /* Test CR0 */ 6735 fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0); 6736 fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1); 6737 test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1); 6738 6739 /* Test CR4 */ 6740 fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0); 6741 fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) & 6742 ~(X86_CR4_SMEP | X86_CR4_SMAP); 6743 test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1); 6744 6745 /* Test CR3 */ 6746 cr3_saved = vmcs_read(HOST_CR3); 6747 for (i = cpuid_maxphyaddr(); i < 64; i++) { 6748 cr3 = cr3_saved | (1ul << i); 6749 vmcs_write(HOST_CR3, cr3); 6750 report_prefix_pushf("HOST_CR3 %lx", cr3); 6751 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 6752 report_prefix_pop(); 6753 } 6754 6755 vmcs_write(HOST_CR3, cr3_saved); 6756 } 6757 6758 static void test_efer_vmlaunch(u32 fld, bool ok) 6759 { 6760 if (fld == HOST_EFER) { 6761 if (ok) 6762 test_vmx_vmlaunch(0); 6763 else 6764 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 6765 } else { 6766 if (ok) { 6767 enter_guest(); 6768 report(vmcs_read(EXI_REASON) == VMX_VMCALL, 6769 "vmlaunch succeeds"); 6770 } else { 6771 enter_guest_with_invalid_guest_state(); 6772 report(vmcs_read(EXI_REASON) == (VMX_ENTRY_FAILURE | VMX_FAIL_STATE), 6773 "vmlaunch fails"); 6774 } 6775 advance_guest_state_test(); 6776 } 6777 } 6778 6779 static void test_efer_one(u32 fld, const char * fld_name, u64 efer, 6780 u32 ctrl_fld, u64 ctrl, 6781 int i, const char *efer_bit_name) 6782 { 6783 bool ok; 6784 6785 ok = true; 6786 if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) { 6787 if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64)) 6788 ok = false; 6789 if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64)) 6790 ok = false; 6791 } 6792 if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) { 6793 /* Check LMA too since CR0.PG is set. */ 6794 if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64)) 6795 ok = false; 6796 if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64)) 6797 ok = false; 6798 } 6799 6800 /* 6801 * Skip the test if it would enter the guest in 32-bit mode. 6802 * Perhaps write the test in assembly and make sure it 6803 * can be run in either mode? 6804 */ 6805 if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64)) 6806 return; 6807 6808 vmcs_write(ctrl_fld, ctrl); 6809 vmcs_write(fld, efer); 6810 report_prefix_pushf("%s %s bit turned %s, controls %s", 6811 fld_name, efer_bit_name, 6812 (i & 1) ? "on" : "off", 6813 (i & 2) ? "on" : "off"); 6814 6815 test_efer_vmlaunch(fld, ok); 6816 report_prefix_pop(); 6817 } 6818 6819 static void test_efer_bit(u32 fld, const char * fld_name, 6820 u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit, 6821 const char *efer_bit_name) 6822 { 6823 u64 efer_saved = vmcs_read(fld); 6824 u32 ctrl_saved = vmcs_read(ctrl_fld); 6825 int i; 6826 6827 for (i = 0; i < 4; i++) { 6828 u64 efer = efer_saved & ~efer_bit; 6829 u64 ctrl = ctrl_saved & ~ctrl_bit; 6830 6831 if (i & 1) 6832 efer |= efer_bit; 6833 if (i & 2) 6834 ctrl |= ctrl_bit; 6835 6836 test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl, 6837 i, efer_bit_name); 6838 } 6839 6840 vmcs_write(ctrl_fld, ctrl_saved); 6841 vmcs_write(fld, efer_saved); 6842 } 6843 6844 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld, 6845 u64 ctrl_bit1, u64 ctrl_bit2) 6846 { 6847 u64 efer_saved = vmcs_read(fld); 6848 u32 ctrl_saved = vmcs_read(ctrl_fld); 6849 u64 efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA)); 6850 u64 i; 6851 u64 efer; 6852 6853 if (cpu_has_efer_nx()) 6854 efer_reserved_bits &= ~EFER_NX; 6855 6856 if (!ctrl_bit1) { 6857 printf("\"Load-IA32-EFER\" exit control not supported\n"); 6858 goto test_entry_exit_mode; 6859 } 6860 6861 report_prefix_pushf("%s %lx", fld_name, efer_saved); 6862 test_efer_vmlaunch(fld, true); 6863 report_prefix_pop(); 6864 6865 /* 6866 * Check reserved bits 6867 */ 6868 vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1); 6869 for (i = 0; i < 64; i++) { 6870 if ((1ull << i) & efer_reserved_bits) { 6871 efer = efer_saved | (1ull << i); 6872 vmcs_write(fld, efer); 6873 report_prefix_pushf("%s %lx", fld_name, efer); 6874 test_efer_vmlaunch(fld, true); 6875 report_prefix_pop(); 6876 } 6877 } 6878 6879 vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1); 6880 for (i = 0; i < 64; i++) { 6881 if ((1ull << i) & efer_reserved_bits) { 6882 efer = efer_saved | (1ull << i); 6883 vmcs_write(fld, efer); 6884 report_prefix_pushf("%s %lx", fld_name, efer); 6885 test_efer_vmlaunch(fld, false); 6886 report_prefix_pop(); 6887 } 6888 } 6889 6890 vmcs_write(ctrl_fld, ctrl_saved); 6891 vmcs_write(fld, efer_saved); 6892 6893 /* 6894 * Check LMA and LME bits 6895 */ 6896 test_efer_bit(fld, fld_name, 6897 ctrl_fld, ctrl_bit1, 6898 EFER_LMA, 6899 "EFER_LMA"); 6900 test_efer_bit(fld, fld_name, 6901 ctrl_fld, ctrl_bit1, 6902 EFER_LME, 6903 "EFER_LME"); 6904 6905 test_entry_exit_mode: 6906 test_efer_bit(fld, fld_name, 6907 ctrl_fld, ctrl_bit2, 6908 EFER_LMA, 6909 "EFER_LMA"); 6910 test_efer_bit(fld, fld_name, 6911 ctrl_fld, ctrl_bit2, 6912 EFER_LME, 6913 "EFER_LME"); 6914 } 6915 6916 /* 6917 * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the 6918 * IA32_EFER MSR must be 0 in the field for that register. In addition, 6919 * the values of the LMA and LME bits in the field must each be that of 6920 * the 'host address-space size' VM-exit control. 6921 * 6922 * [Intel SDM] 6923 */ 6924 static void test_host_efer(void) 6925 { 6926 test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS, 6927 ctrl_exit_rev.clr & EXI_LOAD_EFER, 6928 EXI_HOST_64); 6929 } 6930 6931 /* 6932 * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the 6933 * IA32_EFER MSR must be 0 in the field for that register. In addition, 6934 * the values of the LMA and LME bits in the field must each be that of 6935 * the 'IA32e-mode guest' VM-exit control. 6936 */ 6937 static void test_guest_efer(void) 6938 { 6939 if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) { 6940 printf("\"Load-IA32-EFER\" entry control not supported\n"); 6941 return; 6942 } 6943 6944 vmcs_write(GUEST_EFER, rdmsr(MSR_EFER)); 6945 test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS, 6946 ctrl_enter_rev.clr & ENT_LOAD_EFER, 6947 ENT_GUEST_64); 6948 } 6949 6950 /* 6951 * PAT values higher than 8 are uninteresting since they're likely lumped 6952 * in with "8". We only test values above 8 one bit at a time, 6953 * in order to reduce the number of VM-Entries and keep the runtime reasonable. 6954 */ 6955 #define PAT_VAL_LIMIT 8 6956 6957 static void test_pat(u32 field, const char * field_name, u32 ctrl_field, 6958 u64 ctrl_bit) 6959 { 6960 u32 ctrl_saved = vmcs_read(ctrl_field); 6961 u64 pat_saved = vmcs_read(field); 6962 u64 i, val; 6963 u32 j; 6964 int error; 6965 6966 vmcs_clear_bits(ctrl_field, ctrl_bit); 6967 6968 for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) { 6969 /* Test PAT0..PAT7 fields */ 6970 for (j = 0; j < (i ? 8 : 1); j++) { 6971 val = i << j * 8; 6972 vmcs_write(field, val); 6973 if (field == HOST_PAT) { 6974 report_prefix_pushf("%s %lx", field_name, val); 6975 test_vmx_vmlaunch(0); 6976 report_prefix_pop(); 6977 6978 } else { // GUEST_PAT 6979 enter_guest(); 6980 report_guest_state_test("ENT_LOAD_PAT enabled", 6981 VMX_VMCALL, val, 6982 "GUEST_PAT"); 6983 } 6984 } 6985 } 6986 6987 vmcs_set_bits(ctrl_field, ctrl_bit); 6988 for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) { 6989 /* Test PAT0..PAT7 fields */ 6990 for (j = 0; j < (i ? 8 : 1); j++) { 6991 val = i << j * 8; 6992 vmcs_write(field, val); 6993 6994 if (field == HOST_PAT) { 6995 report_prefix_pushf("%s %lx", field_name, val); 6996 if (i == 0x2 || i == 0x3 || i >= 0x8) 6997 error = 6998 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD; 6999 else 7000 error = 0; 7001 7002 test_vmx_vmlaunch(error); 7003 report_prefix_pop(); 7004 7005 } else { // GUEST_PAT 7006 if (i == 0x2 || i == 0x3 || i >= 0x8) { 7007 enter_guest_with_invalid_guest_state(); 7008 report_guest_state_test("ENT_LOAD_PAT " 7009 "enabled", 7010 VMX_FAIL_STATE | VMX_ENTRY_FAILURE, 7011 val, 7012 "GUEST_PAT"); 7013 } else { 7014 enter_guest(); 7015 report_guest_state_test("ENT_LOAD_PAT " 7016 "enabled", 7017 VMX_VMCALL, 7018 val, 7019 "GUEST_PAT"); 7020 } 7021 } 7022 7023 } 7024 } 7025 7026 vmcs_write(ctrl_field, ctrl_saved); 7027 vmcs_write(field, pat_saved); 7028 } 7029 7030 /* 7031 * If the "load IA32_PAT" VM-exit control is 1, the value of the field 7032 * for the IA32_PAT MSR must be one that could be written by WRMSR 7033 * without fault at CPL 0. Specifically, each of the 8 bytes in the 7034 * field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP), 7035 * 6 (WB), or 7 (UC-). 7036 * 7037 * [Intel SDM] 7038 */ 7039 static void test_load_host_pat(void) 7040 { 7041 /* 7042 * "load IA32_PAT" VM-exit control 7043 */ 7044 if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) { 7045 printf("\"Load-IA32-PAT\" exit control not supported\n"); 7046 return; 7047 } 7048 7049 test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT); 7050 } 7051 7052 union cpuidA_eax { 7053 struct { 7054 unsigned int version_id:8; 7055 unsigned int num_counters_gp:8; 7056 unsigned int bit_width:8; 7057 unsigned int mask_length:8; 7058 } split; 7059 unsigned int full; 7060 }; 7061 7062 union cpuidA_edx { 7063 struct { 7064 unsigned int num_counters_fixed:5; 7065 unsigned int bit_width_fixed:8; 7066 unsigned int reserved:9; 7067 } split; 7068 unsigned int full; 7069 }; 7070 7071 static bool valid_pgc(u64 val) 7072 { 7073 struct cpuid id; 7074 union cpuidA_eax eax; 7075 union cpuidA_edx edx; 7076 u64 mask; 7077 7078 id = cpuid(0xA); 7079 eax.full = id.a; 7080 edx.full = id.d; 7081 mask = ~(((1ull << eax.split.num_counters_gp) - 1) | 7082 (((1ull << edx.split.num_counters_fixed) - 1) << 32)); 7083 7084 return !(val & mask); 7085 } 7086 7087 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host) 7088 { 7089 u32 inst_err; 7090 u64 obs; 7091 bool success; 7092 struct vmx_state_area_test_data *data = &vmx_state_area_test_data; 7093 7094 if (host) { 7095 success = vmlaunch_succeeds(); 7096 obs = rdmsr(data->msr); 7097 if (!success) { 7098 inst_err = vmcs_read(VMX_INST_ERROR); 7099 report(xerror == inst_err, "vmlaunch failed, " 7100 "VMX Inst Error is %d (expected %d)", 7101 inst_err, xerror); 7102 } else { 7103 report(!data->enabled || data->exp == obs, 7104 "Host state is 0x%lx (expected 0x%lx)", 7105 obs, data->exp); 7106 report(success != xfail, "vmlaunch succeeded"); 7107 } 7108 } else { 7109 if (xfail) { 7110 enter_guest_with_invalid_guest_state(); 7111 } else { 7112 enter_guest(); 7113 } 7114 report_guest_state_test("load GUEST_PERF_GLOBAL_CTRL", 7115 xreason, GUEST_PERF_GLOBAL_CTRL, 7116 "GUEST_PERF_GLOBAL_CTRL"); 7117 } 7118 } 7119 7120 /* 7121 * test_load_perf_global_ctrl is a generic function for testing the 7122 * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function 7123 * tests the provided ctrl_val when disabled and enabled. 7124 * 7125 * @nr: VMCS field number corresponding to the host/guest state field 7126 * @name: Name of the above VMCS field for printing in test report 7127 * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control 7128 * @ctrl_val: Bit to set on the ctrl_field 7129 */ 7130 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr, 7131 const char *ctrl_name, u64 ctrl_val) 7132 { 7133 u64 ctrl_saved = vmcs_read(ctrl_nr); 7134 u64 pgc_saved = vmcs_read(nr); 7135 u64 i, val; 7136 bool host = nr == HOST_PERF_GLOBAL_CTRL; 7137 struct vmx_state_area_test_data *data = &vmx_state_area_test_data; 7138 7139 data->msr = MSR_CORE_PERF_GLOBAL_CTRL; 7140 msr_bmp_init(); 7141 vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val); 7142 data->enabled = false; 7143 report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s", 7144 ctrl_name); 7145 7146 for (i = 0; i < 64; i++) { 7147 val = 1ull << i; 7148 vmcs_write(nr, val); 7149 report_prefix_pushf("%s = 0x%lx", name, val); 7150 test_pgc_vmlaunch(0, VMX_VMCALL, false, host); 7151 report_prefix_pop(); 7152 } 7153 report_prefix_pop(); 7154 7155 vmcs_write(ctrl_nr, ctrl_saved | ctrl_val); 7156 data->enabled = true; 7157 report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s", 7158 ctrl_name); 7159 for (i = 0; i < 64; i++) { 7160 val = 1ull << i; 7161 data->exp = val; 7162 vmcs_write(nr, val); 7163 report_prefix_pushf("%s = 0x%lx", name, val); 7164 if (valid_pgc(val)) { 7165 test_pgc_vmlaunch(0, VMX_VMCALL, false, host); 7166 } else { 7167 if (host) 7168 test_pgc_vmlaunch( 7169 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD, 7170 0, 7171 true, 7172 host); 7173 else 7174 test_pgc_vmlaunch( 7175 0, 7176 VMX_ENTRY_FAILURE | VMX_FAIL_STATE, 7177 true, 7178 host); 7179 } 7180 report_prefix_pop(); 7181 } 7182 7183 data->enabled = false; 7184 report_prefix_pop(); 7185 vmcs_write(ctrl_nr, ctrl_saved); 7186 vmcs_write(nr, pgc_saved); 7187 } 7188 7189 static void test_load_host_perf_global_ctrl(void) 7190 { 7191 if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) { 7192 printf("\"load IA32_PERF_GLOBAL_CTRL\" exit control not supported\n"); 7193 return; 7194 } 7195 7196 test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL", 7197 EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF); 7198 } 7199 7200 7201 static void test_load_guest_perf_global_ctrl(void) 7202 { 7203 if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) { 7204 printf("\"load IA32_PERF_GLOBAL_CTRL\" entry control not supported\n"); 7205 return; 7206 } 7207 7208 test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL", 7209 ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF); 7210 } 7211 7212 7213 /* 7214 * test_vmcs_field - test a value for the given VMCS field 7215 * @field: VMCS field 7216 * @field_name: string name of VMCS field 7217 * @bit_start: starting bit 7218 * @bit_end: ending bit 7219 * @val: value that the bit range must or must not contain 7220 * @valid_val: whether value given in 'val' must be valid or not 7221 * @error: expected VMCS error when vmentry fails for an invalid value 7222 */ 7223 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start, 7224 u32 bit_end, u64 val, bool valid_val, u32 error) 7225 { 7226 u64 field_saved = vmcs_read(field); 7227 u32 i; 7228 u64 tmp; 7229 u32 bit_on; 7230 u64 mask = ~0ull; 7231 7232 mask = (mask >> bit_end) << bit_end; 7233 mask = mask | ((1 << bit_start) - 1); 7234 tmp = (field_saved & mask) | (val << bit_start); 7235 7236 vmcs_write(field, tmp); 7237 report_prefix_pushf("%s %lx", field_name, tmp); 7238 if (valid_val) 7239 test_vmx_vmlaunch(0); 7240 else 7241 test_vmx_vmlaunch(error); 7242 report_prefix_pop(); 7243 7244 for (i = bit_start; i <= bit_end; i = i + 2) { 7245 bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1; 7246 if (bit_on) 7247 tmp = field_saved | (1ull << i); 7248 else 7249 tmp = field_saved & ~(1ull << i); 7250 vmcs_write(field, tmp); 7251 report_prefix_pushf("%s %lx", field_name, tmp); 7252 if (valid_val) 7253 test_vmx_vmlaunch(error); 7254 else 7255 test_vmx_vmlaunch(0); 7256 report_prefix_pop(); 7257 } 7258 7259 vmcs_write(field, field_saved); 7260 } 7261 7262 static void test_canonical(u64 field, const char * field_name) 7263 { 7264 u64 addr_saved = vmcs_read(field); 7265 7266 report_prefix_pushf("%s %lx", field_name, addr_saved); 7267 if (is_canonical(addr_saved)) { 7268 test_vmx_vmlaunch(0); 7269 report_prefix_pop(); 7270 7271 vmcs_write(field, NONCANONICAL); 7272 report_prefix_pushf("%s %llx", field_name, NONCANONICAL); 7273 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 7274 7275 vmcs_write(field, addr_saved); 7276 } else { 7277 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 7278 } 7279 report_prefix_pop(); 7280 } 7281 7282 #define TEST_RPL_TI_FLAGS(reg, name) \ 7283 test_vmcs_field(reg, name, 0, 2, 0x0, true, \ 7284 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 7285 7286 #define TEST_CS_TR_FLAGS(reg, name) \ 7287 test_vmcs_field(reg, name, 3, 15, 0x0000, false, \ 7288 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 7289 7290 /* 7291 * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the 7292 * RPL (bits 1:0) and the TI flag (bit 2) must be 0. 7293 * 2. The selector fields for CS and TR cannot be 0000H. 7294 * 3. The selector field for SS cannot be 0000H if the "host address-space 7295 * size" VM-exit control is 0. 7296 * 4. On processors that support Intel 64 architecture, the base-address 7297 * fields for FS, GS and TR must contain canonical addresses. 7298 */ 7299 static void test_host_segment_regs(void) 7300 { 7301 u16 selector_saved; 7302 7303 /* 7304 * Test RPL and TI flags 7305 */ 7306 TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS"); 7307 TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS"); 7308 TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS"); 7309 TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES"); 7310 TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS"); 7311 TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS"); 7312 TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR"); 7313 7314 /* 7315 * Test that CS and TR fields can not be 0x0000 7316 */ 7317 TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS"); 7318 TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR"); 7319 7320 /* 7321 * SS field can not be 0x0000 if "host address-space size" VM-exit 7322 * control is 0 7323 */ 7324 selector_saved = vmcs_read(HOST_SEL_SS); 7325 vmcs_write(HOST_SEL_SS, 0); 7326 report_prefix_pushf("HOST_SEL_SS 0"); 7327 if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) { 7328 test_vmx_vmlaunch(0); 7329 } else { 7330 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 7331 } 7332 report_prefix_pop(); 7333 7334 vmcs_write(HOST_SEL_SS, selector_saved); 7335 7336 #ifdef __x86_64__ 7337 /* 7338 * Base address for FS, GS and TR must be canonical 7339 */ 7340 test_canonical(HOST_BASE_FS, "HOST_BASE_FS"); 7341 test_canonical(HOST_BASE_GS, "HOST_BASE_GS"); 7342 test_canonical(HOST_BASE_TR, "HOST_BASE_TR"); 7343 #endif 7344 } 7345 7346 /* 7347 * On processors that support Intel 64 architecture, the base-address 7348 * fields for GDTR and IDTR must contain canonical addresses. 7349 */ 7350 static void test_host_desc_tables(void) 7351 { 7352 #ifdef __x86_64__ 7353 test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR"); 7354 test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR"); 7355 #endif 7356 } 7357 7358 /* 7359 * If the "host address-space size" VM-exit control is 0, the following must 7360 * hold: 7361 * - The "IA-32e mode guest" VM-entry control is 0. 7362 * - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0. 7363 * - Bits 63:32 in the RIP field are 0. 7364 * 7365 * If the "host address-space size" VM-exit control is 1, the following must 7366 * hold: 7367 * - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1. 7368 * - The RIP field contains a canonical address. 7369 * 7370 */ 7371 static void test_host_addr_size(void) 7372 { 7373 u64 cr4_saved = vmcs_read(HOST_CR4); 7374 u64 rip_saved = vmcs_read(HOST_RIP); 7375 u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS); 7376 int i; 7377 u64 tmp; 7378 7379 if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) { 7380 vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64); 7381 report_prefix_pushf("\"IA-32e mode guest\" enabled"); 7382 test_vmx_vmlaunch(0); 7383 report_prefix_pop(); 7384 7385 vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE); 7386 report_prefix_pushf("\"CR4.PCIDE\" set"); 7387 test_vmx_vmlaunch(0); 7388 report_prefix_pop(); 7389 7390 for (i = 32; i <= 63; i = i + 4) { 7391 tmp = rip_saved | 1ull << i; 7392 vmcs_write(HOST_RIP, tmp); 7393 report_prefix_pushf("HOST_RIP %lx", tmp); 7394 test_vmx_vmlaunch(0); 7395 report_prefix_pop(); 7396 } 7397 7398 if (cr4_saved & X86_CR4_PAE) { 7399 vmcs_write(HOST_CR4, cr4_saved & ~X86_CR4_PAE); 7400 report_prefix_pushf("\"CR4.PAE\" unset"); 7401 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 7402 } else { 7403 report_prefix_pushf("\"CR4.PAE\" set"); 7404 test_vmx_vmlaunch(0); 7405 } 7406 report_prefix_pop(); 7407 7408 vmcs_write(HOST_RIP, NONCANONICAL); 7409 report_prefix_pushf("HOST_RIP %llx", NONCANONICAL); 7410 test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD); 7411 report_prefix_pop(); 7412 7413 vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64); 7414 vmcs_write(HOST_RIP, rip_saved); 7415 vmcs_write(HOST_CR4, cr4_saved); 7416 } 7417 } 7418 7419 /* 7420 * Check that the virtual CPU checks the VMX Host State Area as 7421 * documented in the Intel SDM. 7422 */ 7423 static void vmx_host_state_area_test(void) 7424 { 7425 /* 7426 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will 7427 * fail due to invalid guest state, should we make it that 7428 * far. 7429 */ 7430 vmcs_write(GUEST_RFLAGS, 0); 7431 7432 test_host_ctl_regs(); 7433 7434 test_sysenter_field(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP"); 7435 test_sysenter_field(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP"); 7436 7437 test_host_efer(); 7438 test_load_host_pat(); 7439 test_host_segment_regs(); 7440 test_host_desc_tables(); 7441 test_host_addr_size(); 7442 test_load_host_perf_global_ctrl(); 7443 } 7444 7445 /* 7446 * If the "load IA32_PAT" VM-entry control is 1, the value of the field 7447 * for the IA32_PAT MSR must be one that could be written by WRMSR 7448 * without fault at CPL 0. Specifically, each of the 8 bytes in the 7449 * field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP), 7450 * 6 (WB), or 7 (UC-). 7451 * 7452 * [Intel SDM] 7453 */ 7454 static void test_load_guest_pat(void) 7455 { 7456 /* 7457 * "load IA32_PAT" VM-entry control 7458 */ 7459 if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) { 7460 printf("\"Load-IA32-PAT\" entry control not supported\n"); 7461 return; 7462 } 7463 7464 test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT); 7465 } 7466 7467 /* 7468 * Check that the virtual CPU checks the VMX Guest State Area as 7469 * documented in the Intel SDM. 7470 */ 7471 static void vmx_guest_state_area_test(void) 7472 { 7473 vmx_set_test_stage(1); 7474 test_set_guest(guest_state_test_main); 7475 7476 test_load_guest_pat(); 7477 test_guest_efer(); 7478 test_load_guest_perf_global_ctrl(); 7479 7480 /* 7481 * Let the guest finish execution 7482 */ 7483 vmx_set_test_stage(2); 7484 enter_guest(); 7485 } 7486 7487 static bool valid_vmcs_for_vmentry(void) 7488 { 7489 struct vmcs *current_vmcs = NULL; 7490 7491 if (vmcs_save(¤t_vmcs)) 7492 return false; 7493 7494 return current_vmcs && !current_vmcs->hdr.shadow_vmcs; 7495 } 7496 7497 static void try_vmentry_in_movss_shadow(void) 7498 { 7499 u32 vm_inst_err; 7500 u32 flags; 7501 bool early_failure = false; 7502 u32 expected_flags = X86_EFLAGS_FIXED; 7503 bool valid_vmcs = valid_vmcs_for_vmentry(); 7504 7505 expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF; 7506 7507 /* 7508 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to 7509 * unsupported VMCS component"). 7510 */ 7511 vmcs_write(~0u, 0); 7512 7513 __asm__ __volatile__ ("mov %[host_rsp], %%edx;" 7514 "vmwrite %%rsp, %%rdx;" 7515 "mov 0f, %%rax;" 7516 "mov %[host_rip], %%edx;" 7517 "vmwrite %%rax, %%rdx;" 7518 "mov $-1, %%ah;" 7519 "sahf;" 7520 "mov %%ss, %%ax;" 7521 "mov %%ax, %%ss;" 7522 "vmlaunch;" 7523 "mov $1, %[early_failure];" 7524 "0: lahf;" 7525 "movzbl %%ah, %[flags]" 7526 : [early_failure] "+r" (early_failure), 7527 [flags] "=&a" (flags) 7528 : [host_rsp] "i" (HOST_RSP), 7529 [host_rip] "i" (HOST_RIP) 7530 : "rdx", "cc", "memory"); 7531 vm_inst_err = vmcs_read(VMX_INST_ERROR); 7532 7533 report(early_failure, "Early VM-entry failure"); 7534 report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)", 7535 expected_flags, flags); 7536 if (valid_vmcs) 7537 report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, 7538 "VM-instruction error is %d (actual %d)", 7539 VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err); 7540 } 7541 7542 static void vmentry_movss_shadow_test(void) 7543 { 7544 struct vmcs *orig_vmcs; 7545 7546 TEST_ASSERT(!vmcs_save(&orig_vmcs)); 7547 7548 /* 7549 * Set the launched flag on the current VMCS to verify the correct 7550 * error priority, below. 7551 */ 7552 test_set_guest(v2_null_test_guest); 7553 enter_guest(); 7554 7555 /* 7556 * With bit 1 of the guest's RFLAGS clear, VM-entry should 7557 * fail due to invalid guest state (if we make it that far). 7558 */ 7559 vmcs_write(GUEST_RFLAGS, 0); 7560 7561 /* 7562 * "VM entry with events blocked by MOV SS" takes precedence over 7563 * "VMLAUNCH with non-clear VMCS." 7564 */ 7565 report_prefix_push("valid current-VMCS"); 7566 try_vmentry_in_movss_shadow(); 7567 report_prefix_pop(); 7568 7569 /* 7570 * VMfailInvalid takes precedence over "VM entry with events 7571 * blocked by MOV SS." 7572 */ 7573 TEST_ASSERT(!vmcs_clear(orig_vmcs)); 7574 report_prefix_push("no current-VMCS"); 7575 try_vmentry_in_movss_shadow(); 7576 report_prefix_pop(); 7577 7578 TEST_ASSERT(!make_vmcs_current(orig_vmcs)); 7579 vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED); 7580 } 7581 7582 static void vmx_cr_load_test(void) 7583 { 7584 unsigned long cr3, cr4, orig_cr3, orig_cr4; 7585 7586 orig_cr4 = read_cr4(); 7587 orig_cr3 = read_cr3(); 7588 7589 if (!this_cpu_has(X86_FEATURE_PCID)) { 7590 report_skip("PCID not detected"); 7591 return; 7592 } 7593 if (!this_cpu_has(X86_FEATURE_MCE)) { 7594 report_skip("MCE not detected"); 7595 return; 7596 } 7597 7598 TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK)); 7599 7600 /* Enable PCID for L1. */ 7601 cr4 = orig_cr4 | X86_CR4_PCIDE; 7602 cr3 = orig_cr3 | 0x1; 7603 TEST_ASSERT(!write_cr4_checking(cr4)); 7604 write_cr3(cr3); 7605 7606 test_set_guest(v2_null_test_guest); 7607 vmcs_write(HOST_CR4, cr4); 7608 vmcs_write(HOST_CR3, cr3); 7609 enter_guest(); 7610 7611 /* 7612 * No exception is expected. 7613 * 7614 * NB. KVM loads the last guest write to CR4 into CR4 read 7615 * shadow. In order to trigger an exit to KVM, we can toggle a 7616 * bit that is owned by KVM. We use CR4.MCE, which shall 7617 * have no side effect because normally no guest MCE (e.g., as the 7618 * result of bad memory) would happen during this test. 7619 */ 7620 TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE)); 7621 7622 /* Cleanup L1 state. */ 7623 write_cr3(orig_cr3); 7624 TEST_ASSERT(!write_cr4_checking(orig_cr4)); 7625 } 7626 7627 static void vmx_nm_test_guest(void) 7628 { 7629 write_cr0(read_cr0() | X86_CR0_TS); 7630 asm volatile("fnop"); 7631 } 7632 7633 static void check_nm_exit(const char *test) 7634 { 7635 u32 reason = vmcs_read(EXI_REASON); 7636 u32 intr_info = vmcs_read(EXI_INTR_INFO); 7637 const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | 7638 NM_VECTOR; 7639 7640 report(reason == VMX_EXC_NMI && intr_info == expected, "%s", test); 7641 } 7642 7643 /* 7644 * This test checks that: 7645 * 7646 * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then 7647 * a subsequent #NM VM-exit is reflected to L1. 7648 * 7649 * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a 7650 * subsequent #NM VM-exit is reflected to L1. 7651 */ 7652 static void vmx_nm_test(void) 7653 { 7654 unsigned long cr0 = read_cr0(); 7655 7656 test_set_guest(vmx_nm_test_guest); 7657 7658 /* 7659 * L1 wants to intercept #NM exceptions encountered in L2. 7660 */ 7661 vmcs_write(EXC_BITMAP, 1 << NM_VECTOR); 7662 7663 /* 7664 * Launch L2 with CR0.TS clear, but don't claim host ownership of 7665 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop, 7666 * which will raise #NM. L0 should reflect the #NM VM-exit to L1. 7667 */ 7668 vmcs_write(CR0_MASK, 0); 7669 vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS); 7670 enter_guest(); 7671 check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1"); 7672 7673 /* 7674 * Re-enter L2 at the fnop instruction, with CR0.TS clear but 7675 * CR0.EM set. The fnop will still raise #NM, and L0 should 7676 * reflect the #NM VM-exit to L1. 7677 */ 7678 vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM); 7679 enter_guest(); 7680 check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1"); 7681 7682 /* 7683 * Re-enter L2 at the fnop instruction, with both CR0.TS and 7684 * CR0.EM clear. There will be no #NM, and the L2 guest should 7685 * exit normally. 7686 */ 7687 vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM)); 7688 enter_guest(); 7689 } 7690 7691 bool vmx_pending_event_ipi_fired; 7692 static void vmx_pending_event_ipi_isr(isr_regs_t *regs) 7693 { 7694 vmx_pending_event_ipi_fired = true; 7695 eoi(); 7696 } 7697 7698 bool vmx_pending_event_guest_run; 7699 static void vmx_pending_event_guest(void) 7700 { 7701 vmcall(); 7702 vmx_pending_event_guest_run = true; 7703 } 7704 7705 static void vmx_pending_event_test_core(bool guest_hlt) 7706 { 7707 int ipi_vector = 0xf1; 7708 7709 vmx_pending_event_ipi_fired = false; 7710 handle_irq(ipi_vector, vmx_pending_event_ipi_isr); 7711 7712 vmx_pending_event_guest_run = false; 7713 test_set_guest(vmx_pending_event_guest); 7714 7715 vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT); 7716 7717 enter_guest(); 7718 skip_exit_vmcall(); 7719 7720 if (guest_hlt) 7721 vmcs_write(GUEST_ACTV_STATE, ACTV_HLT); 7722 7723 irq_disable(); 7724 apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | 7725 APIC_DM_FIXED | ipi_vector, 7726 0); 7727 7728 enter_guest(); 7729 7730 assert_exit_reason(VMX_EXTINT); 7731 report(!vmx_pending_event_guest_run, 7732 "Guest did not run before host received IPI"); 7733 7734 irq_enable(); 7735 asm volatile ("nop"); 7736 irq_disable(); 7737 report(vmx_pending_event_ipi_fired, 7738 "Got pending interrupt after IRQ enabled"); 7739 7740 if (guest_hlt) 7741 vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE); 7742 7743 enter_guest(); 7744 report(vmx_pending_event_guest_run, 7745 "Guest finished running when no interrupt"); 7746 } 7747 7748 static void vmx_pending_event_test(void) 7749 { 7750 vmx_pending_event_test_core(false); 7751 } 7752 7753 static void vmx_pending_event_hlt_test(void) 7754 { 7755 vmx_pending_event_test_core(true); 7756 } 7757 7758 static int vmx_window_test_db_count; 7759 7760 static void vmx_window_test_db_handler(struct ex_regs *regs) 7761 { 7762 vmx_window_test_db_count++; 7763 } 7764 7765 static void vmx_nmi_window_test_guest(void) 7766 { 7767 handle_exception(DB_VECTOR, vmx_window_test_db_handler); 7768 7769 asm volatile("vmcall\n\t" 7770 "nop\n\t"); 7771 7772 handle_exception(DB_VECTOR, NULL); 7773 } 7774 7775 static void verify_nmi_window_exit(u64 rip) 7776 { 7777 u32 exit_reason = vmcs_read(EXI_REASON); 7778 7779 report(exit_reason == VMX_NMI_WINDOW, 7780 "Exit reason (%d) is 'NMI window'", exit_reason); 7781 report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx", 7782 vmcs_read(GUEST_RIP), rip); 7783 vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE); 7784 } 7785 7786 static void vmx_nmi_window_test(void) 7787 { 7788 u64 nop_addr; 7789 void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]); 7790 7791 if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) { 7792 report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control."); 7793 return; 7794 } 7795 7796 if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) { 7797 report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control."); 7798 return; 7799 } 7800 7801 vmx_window_test_db_count = 0; 7802 7803 report_prefix_push("NMI-window"); 7804 test_set_guest(vmx_nmi_window_test_guest); 7805 vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI); 7806 enter_guest(); 7807 skip_exit_vmcall(); 7808 nop_addr = vmcs_read(GUEST_RIP); 7809 7810 /* 7811 * Ask for "NMI-window exiting," and expect an immediate VM-exit. 7812 * RIP will not advance. 7813 */ 7814 report_prefix_push("active, no blocking"); 7815 vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW); 7816 enter_guest(); 7817 verify_nmi_window_exit(nop_addr); 7818 report_prefix_pop(); 7819 7820 /* 7821 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect 7822 * a VM-exit on the next instruction after the nop. (The nop 7823 * is one byte.) 7824 */ 7825 report_prefix_push("active, blocking by MOV-SS"); 7826 vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS); 7827 enter_guest(); 7828 verify_nmi_window_exit(nop_addr + 1); 7829 report_prefix_pop(); 7830 7831 /* 7832 * Ask for "NMI-window exiting" (with event injection), and 7833 * expect a VM-exit after the event is injected. (RIP should 7834 * be at the address specified in the IDT entry for #DB.) 7835 */ 7836 report_prefix_push("active, no blocking, injecting #DB"); 7837 vmcs_write(ENT_INTR_INFO, 7838 INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR); 7839 enter_guest(); 7840 verify_nmi_window_exit((u64)db_fault_addr); 7841 report_prefix_pop(); 7842 7843 /* 7844 * Ask for "NMI-window exiting" with NMI blocking, and expect 7845 * a VM-exit after the next IRET (i.e. after the #DB handler 7846 * returns). So, RIP should be back at one byte past the nop. 7847 */ 7848 report_prefix_push("active, blocking by NMI"); 7849 vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI); 7850 enter_guest(); 7851 verify_nmi_window_exit(nop_addr + 1); 7852 report(vmx_window_test_db_count == 1, 7853 "#DB handler executed once (actual %d times)", 7854 vmx_window_test_db_count); 7855 report_prefix_pop(); 7856 7857 if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) { 7858 report_skip("CPU does not support activity state HLT."); 7859 } else { 7860 /* 7861 * Ask for "NMI-window exiting" when entering activity 7862 * state HLT, and expect an immediate VM-exit. RIP is 7863 * still one byte past the nop. 7864 */ 7865 report_prefix_push("halted, no blocking"); 7866 vmcs_write(GUEST_ACTV_STATE, ACTV_HLT); 7867 enter_guest(); 7868 verify_nmi_window_exit(nop_addr + 1); 7869 report_prefix_pop(); 7870 7871 /* 7872 * Ask for "NMI-window exiting" when entering activity 7873 * state HLT (with event injection), and expect a 7874 * VM-exit after the event is injected. (RIP should be 7875 * at the address specified in the IDT entry for #DB.) 7876 */ 7877 report_prefix_push("halted, no blocking, injecting #DB"); 7878 vmcs_write(GUEST_ACTV_STATE, ACTV_HLT); 7879 vmcs_write(ENT_INTR_INFO, 7880 INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | 7881 DB_VECTOR); 7882 enter_guest(); 7883 verify_nmi_window_exit((u64)db_fault_addr); 7884 report_prefix_pop(); 7885 } 7886 7887 vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW); 7888 enter_guest(); 7889 report_prefix_pop(); 7890 } 7891 7892 static void vmx_intr_window_test_guest(void) 7893 { 7894 handle_exception(DB_VECTOR, vmx_window_test_db_handler); 7895 7896 /* 7897 * The two consecutive STIs are to ensure that only the first 7898 * one has a shadow. Note that NOP and STI are one byte 7899 * instructions. 7900 */ 7901 asm volatile("vmcall\n\t" 7902 "nop\n\t" 7903 "sti\n\t" 7904 "sti\n\t"); 7905 7906 handle_exception(DB_VECTOR, NULL); 7907 } 7908 7909 static void verify_intr_window_exit(u64 rip) 7910 { 7911 u32 exit_reason = vmcs_read(EXI_REASON); 7912 7913 report(exit_reason == VMX_INTR_WINDOW, 7914 "Exit reason (%d) is 'interrupt window'", exit_reason); 7915 report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx", 7916 vmcs_read(GUEST_RIP), rip); 7917 vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE); 7918 } 7919 7920 static void vmx_intr_window_test(void) 7921 { 7922 u64 vmcall_addr; 7923 u64 nop_addr; 7924 unsigned int orig_db_gate_type; 7925 void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]); 7926 7927 if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) { 7928 report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control."); 7929 return; 7930 } 7931 7932 /* 7933 * Change the IDT entry for #DB from interrupt gate to trap gate, 7934 * so that it won't clear RFLAGS.IF. We don't want interrupts to 7935 * be disabled after vectoring a #DB. 7936 */ 7937 orig_db_gate_type = boot_idt[DB_VECTOR].type; 7938 boot_idt[DB_VECTOR].type = 15; 7939 7940 report_prefix_push("interrupt-window"); 7941 test_set_guest(vmx_intr_window_test_guest); 7942 enter_guest(); 7943 assert_exit_reason(VMX_VMCALL); 7944 vmcall_addr = vmcs_read(GUEST_RIP); 7945 7946 /* 7947 * Ask for "interrupt-window exiting" with RFLAGS.IF set and 7948 * no blocking; expect an immediate VM-exit. Note that we have 7949 * not advanced past the vmcall instruction yet, so RIP should 7950 * point to the vmcall instruction. 7951 */ 7952 report_prefix_push("active, no blocking, RFLAGS.IF=1"); 7953 vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW); 7954 vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF); 7955 enter_guest(); 7956 verify_intr_window_exit(vmcall_addr); 7957 report_prefix_pop(); 7958 7959 /* 7960 * Ask for "interrupt-window exiting" (with event injection) 7961 * with RFLAGS.IF set and no blocking; expect a VM-exit after 7962 * the event is injected. That is, RIP should should be at the 7963 * address specified in the IDT entry for #DB. 7964 */ 7965 report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB"); 7966 vmcs_write(ENT_INTR_INFO, 7967 INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR); 7968 vmcall_addr = vmcs_read(GUEST_RIP); 7969 enter_guest(); 7970 verify_intr_window_exit((u64)db_fault_addr); 7971 report_prefix_pop(); 7972 7973 /* 7974 * Let the L2 guest run through the IRET, back to the VMCALL. 7975 * We have to clear the "interrupt-window exiting" 7976 * VM-execution control, or it would just keep causing 7977 * VM-exits. Then, advance past the VMCALL and set the 7978 * "interrupt-window exiting" VM-execution control again. 7979 */ 7980 vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW); 7981 enter_guest(); 7982 skip_exit_vmcall(); 7983 nop_addr = vmcs_read(GUEST_RIP); 7984 vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW); 7985 7986 /* 7987 * Ask for "interrupt-window exiting" in a MOV-SS shadow with 7988 * RFLAGS.IF set, and expect a VM-exit on the next 7989 * instruction. (NOP is one byte.) 7990 */ 7991 report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1"); 7992 vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS); 7993 enter_guest(); 7994 verify_intr_window_exit(nop_addr + 1); 7995 report_prefix_pop(); 7996 7997 /* 7998 * Back up to the NOP and ask for "interrupt-window exiting" 7999 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit 8000 * on the next instruction. (NOP is one byte.) 8001 */ 8002 report_prefix_push("active, blocking by STI, RFLAGS.IF=1"); 8003 vmcs_write(GUEST_RIP, nop_addr); 8004 vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI); 8005 enter_guest(); 8006 verify_intr_window_exit(nop_addr + 1); 8007 report_prefix_pop(); 8008 8009 /* 8010 * Ask for "interrupt-window exiting" with RFLAGS.IF clear, 8011 * and expect a VM-exit on the instruction following the STI 8012 * shadow. Only the first STI (which is one byte past the NOP) 8013 * should have a shadow. The second STI (which is two bytes 8014 * past the NOP) has no shadow. Therefore, the interrupt 8015 * window opens at three bytes past the NOP. 8016 */ 8017 report_prefix_push("active, RFLAGS.IF = 0"); 8018 vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED); 8019 enter_guest(); 8020 verify_intr_window_exit(nop_addr + 3); 8021 report_prefix_pop(); 8022 8023 if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) { 8024 report_skip("CPU does not support activity state HLT."); 8025 } else { 8026 /* 8027 * Ask for "interrupt-window exiting" when entering 8028 * activity state HLT, and expect an immediate 8029 * VM-exit. RIP is still three bytes past the nop. 8030 */ 8031 report_prefix_push("halted, no blocking"); 8032 vmcs_write(GUEST_ACTV_STATE, ACTV_HLT); 8033 enter_guest(); 8034 verify_intr_window_exit(nop_addr + 3); 8035 report_prefix_pop(); 8036 8037 /* 8038 * Ask for "interrupt-window exiting" when entering 8039 * activity state HLT (with event injection), and 8040 * expect a VM-exit after the event is injected. That 8041 * is, RIP should should be at the address specified 8042 * in the IDT entry for #DB. 8043 */ 8044 report_prefix_push("halted, no blocking, injecting #DB"); 8045 vmcs_write(GUEST_ACTV_STATE, ACTV_HLT); 8046 vmcs_write(ENT_INTR_INFO, 8047 INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | 8048 DB_VECTOR); 8049 enter_guest(); 8050 verify_intr_window_exit((u64)db_fault_addr); 8051 report_prefix_pop(); 8052 } 8053 8054 boot_idt[DB_VECTOR].type = orig_db_gate_type; 8055 vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW); 8056 enter_guest(); 8057 report_prefix_pop(); 8058 } 8059 8060 #define GUEST_TSC_OFFSET (1u << 30) 8061 8062 static u64 guest_tsc; 8063 8064 static void vmx_store_tsc_test_guest(void) 8065 { 8066 guest_tsc = rdtsc(); 8067 } 8068 8069 /* 8070 * This test ensures that when IA32_TSC is in the VM-exit MSR-store 8071 * list, the value saved is not subject to the TSC offset that is 8072 * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution. 8073 */ 8074 static void vmx_store_tsc_test(void) 8075 { 8076 struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC }; 8077 u64 low, high; 8078 8079 if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) { 8080 report_skip("'Use TSC offsetting' not supported"); 8081 return; 8082 } 8083 8084 test_set_guest(vmx_store_tsc_test_guest); 8085 8086 vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET); 8087 vmcs_write(EXI_MSR_ST_CNT, 1); 8088 vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry)); 8089 vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET); 8090 8091 low = rdtsc(); 8092 enter_guest(); 8093 high = rdtsc(); 8094 8095 report(low + GUEST_TSC_OFFSET <= guest_tsc && 8096 guest_tsc <= high + GUEST_TSC_OFFSET, 8097 "RDTSC value in the guest (%lu) is in range [%lu, %lu]", 8098 guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET); 8099 report(low <= msr_entry.value && msr_entry.value <= high, 8100 "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]", 8101 msr_entry.value, low, high); 8102 } 8103 8104 static void vmx_db_test_guest(void) 8105 { 8106 /* 8107 * For a hardware generated single-step #DB. 8108 */ 8109 asm volatile("vmcall;" 8110 "nop;" 8111 ".Lpost_nop:"); 8112 /* 8113 * ...in a MOVSS shadow, with pending debug exceptions. 8114 */ 8115 asm volatile("vmcall;" 8116 "nop;" 8117 ".Lpost_movss_nop:"); 8118 /* 8119 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and 8120 * emulates it in software.) 8121 */ 8122 asm volatile("vmcall;" 8123 "wbinvd;" 8124 ".Lpost_wbinvd:"); 8125 /* 8126 * ...in a MOVSS shadow, with pending debug exceptions. 8127 */ 8128 asm volatile("vmcall;" 8129 "wbinvd;" 8130 ".Lpost_movss_wbinvd:"); 8131 /* 8132 * For a hardware generated single-step #DB in a transactional region. 8133 */ 8134 asm volatile("vmcall;" 8135 ".Lxbegin: xbegin .Lskip_rtm;" 8136 "xend;" 8137 ".Lskip_rtm:"); 8138 } 8139 8140 /* 8141 * Clear the pending debug exceptions and RFLAGS.TF and re-enter 8142 * L2. No #DB is delivered and L2 continues to the next point of 8143 * interest. 8144 */ 8145 static void dismiss_db(void) 8146 { 8147 vmcs_write(GUEST_PENDING_DEBUG, 0); 8148 vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED); 8149 enter_guest(); 8150 } 8151 8152 /* 8153 * Check a variety of VMCS fields relevant to an intercepted #DB exception. 8154 * Then throw away the #DB exception and resume L2. 8155 */ 8156 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg, 8157 void *expected_rip, u64 expected_exit_qual, 8158 u64 expected_dr6) 8159 { 8160 u32 reason = vmcs_read(EXI_REASON); 8161 u32 intr_info = vmcs_read(EXI_INTR_INFO); 8162 u64 exit_qual = vmcs_read(EXI_QUALIFICATION); 8163 u64 guest_rip = vmcs_read(GUEST_RIP); 8164 u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG); 8165 u64 dr6 = read_dr6(); 8166 const u32 expected_intr_info = INTR_INFO_VALID_MASK | 8167 INTR_TYPE_HARD_EXCEPTION | DB_VECTOR; 8168 8169 report(reason == VMX_EXC_NMI && intr_info == expected_intr_info, 8170 "Expected #DB VM-exit"); 8171 report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)", 8172 expected_rip, guest_rip); 8173 report_xfail(xfail_pdbg, 0 == guest_pending_dbg, 8174 "Expected pending debug exceptions 0 (actual %lx)", 8175 guest_pending_dbg); 8176 report_xfail(xfail_qual, expected_exit_qual == exit_qual, 8177 "Expected exit qualification %lx (actual %lx)", 8178 expected_exit_qual, exit_qual); 8179 report_xfail(xfail_dr6, expected_dr6 == dr6, 8180 "Expected DR6 %lx (actual %lx)", expected_dr6, dr6); 8181 dismiss_db(); 8182 } 8183 8184 /* 8185 * Assuming the guest has just exited on a VMCALL instruction, skip 8186 * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If 8187 * pending debug exceptions are non-zero, set the VMCS up as if the 8188 * previous instruction was a MOVSS that generated the indicated 8189 * pending debug exceptions. Then enter L2. 8190 */ 8191 static void single_step_guest(const char *test_name, u64 starting_dr6, 8192 u64 pending_debug_exceptions) 8193 { 8194 printf("\n%s\n", test_name); 8195 skip_exit_vmcall(); 8196 write_dr6(starting_dr6); 8197 vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF); 8198 if (pending_debug_exceptions) { 8199 vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions); 8200 vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS); 8201 } 8202 enter_guest(); 8203 } 8204 8205 /* 8206 * When L1 intercepts #DB, verify that a single-step trap clears 8207 * pending debug exceptions, populates the exit qualification field 8208 * properly, and that DR6 is not prematurely clobbered. In a 8209 * (simulated) MOVSS shadow, make sure that the pending debug 8210 * exception bits are properly accumulated into the exit qualification 8211 * field. 8212 */ 8213 static void vmx_db_test(void) 8214 { 8215 /* 8216 * We are going to set a few arbitrary bits in DR6 to verify that 8217 * (a) DR6 is not modified by an intercepted #DB, and 8218 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into 8219 * the exit qualification field for a subsequent #DB exception. 8220 */ 8221 const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1; 8222 extern char post_nop asm(".Lpost_nop"); 8223 extern char post_movss_nop asm(".Lpost_movss_nop"); 8224 extern char post_wbinvd asm(".Lpost_wbinvd"); 8225 extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd"); 8226 extern char xbegin asm(".Lxbegin"); 8227 extern char skip_rtm asm(".Lskip_rtm"); 8228 8229 /* 8230 * L1 wants to intercept #DB exceptions encountered in L2. 8231 */ 8232 vmcs_write(EXC_BITMAP, BIT(DB_VECTOR)); 8233 8234 /* 8235 * Start L2 and run it up to the first point of interest. 8236 */ 8237 test_set_guest(vmx_db_test_guest); 8238 enter_guest(); 8239 8240 /* 8241 * Hardware-delivered #DB trap for single-step sets the 8242 * standard that L0 has to follow for emulated instructions. 8243 */ 8244 single_step_guest("Hardware delivered single-step", starting_dr6, 0); 8245 check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6); 8246 8247 /* 8248 * Hardware-delivered #DB trap for single-step in MOVSS shadow 8249 * also sets the standard that L0 has to follow for emulated 8250 * instructions. Here, we establish the VMCS pending debug 8251 * exceptions to indicate that the simulated MOVSS triggered a 8252 * data breakpoint as well as the single-step trap. 8253 */ 8254 single_step_guest("Hardware delivered single-step in MOVSS shadow", 8255 starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 ); 8256 check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0, 8257 starting_dr6); 8258 8259 /* 8260 * L0 synthesized #DB trap for single-step is buggy, because 8261 * kvm (a) clobbers DR6 too early, and (b) tries its best to 8262 * reconstitute the exit qualification from the prematurely 8263 * modified DR6, but fails miserably. 8264 */ 8265 single_step_guest("Software synthesized single-step", starting_dr6, 0); 8266 check_db_exit(true, true, false, &post_wbinvd, DR_STEP, starting_dr6); 8267 8268 /* 8269 * L0 synthesized #DB trap for single-step in MOVSS shadow is 8270 * even worse, because L0 also leaves the pending debug 8271 * exceptions in the VMCS instead of accumulating them into 8272 * the exit qualification field for the #DB exception. 8273 */ 8274 single_step_guest("Software synthesized single-step in MOVSS shadow", 8275 starting_dr6, BIT(12) | DR_STEP | DR_TRAP0); 8276 check_db_exit(true, true, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0, 8277 starting_dr6); 8278 8279 /* 8280 * Optional RTM test for hardware that supports RTM, to 8281 * demonstrate that the current volume 3 of the SDM 8282 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit 8283 * qualification for debug exceptions is not reserved. It is 8284 * set to 1 if a debug exception (#DB) or a breakpoint 8285 * exception (#BP) occurs inside an RTM region while advanced 8286 * debugging of RTM transactional regions is enabled. 8287 */ 8288 if (this_cpu_has(X86_FEATURE_RTM)) { 8289 vmcs_write(ENT_CONTROLS, 8290 vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS); 8291 /* 8292 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15] 8293 * in the guest to enable advanced debugging of RTM 8294 * transactional regions. 8295 */ 8296 vmcs_write(GUEST_DR7, BIT(11)); 8297 vmcs_write(GUEST_DEBUGCTL, BIT(15)); 8298 single_step_guest("Hardware delivered single-step in " 8299 "transactional region", starting_dr6, 0); 8300 check_db_exit(false, false, false, &xbegin, BIT(16), 8301 starting_dr6); 8302 } else { 8303 vmcs_write(GUEST_RIP, (u64)&skip_rtm); 8304 enter_guest(); 8305 } 8306 } 8307 8308 static void enable_vid(void) 8309 { 8310 void *virtual_apic_page; 8311 8312 assert(cpu_has_apicv()); 8313 8314 disable_intercept_for_x2apic_msrs(); 8315 8316 virtual_apic_page = alloc_page(); 8317 vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page); 8318 8319 vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT); 8320 8321 vmcs_write(EOI_EXIT_BITMAP0, 0x0); 8322 vmcs_write(EOI_EXIT_BITMAP1, 0x0); 8323 vmcs_write(EOI_EXIT_BITMAP2, 0x0); 8324 vmcs_write(EOI_EXIT_BITMAP3, 0x0); 8325 8326 vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW); 8327 vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC); 8328 } 8329 8330 static void trigger_ioapic_scan_thread(void *data) 8331 { 8332 /* Wait until other CPU entered L2 */ 8333 while (vmx_get_test_stage() != 1) 8334 ; 8335 8336 /* Trigger ioapic scan */ 8337 ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL); 8338 vmx_set_test_stage(2); 8339 } 8340 8341 static void irq_79_handler_guest(isr_regs_t *regs) 8342 { 8343 eoi(); 8344 8345 /* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */ 8346 vmcall(); 8347 } 8348 8349 /* 8350 * Constant for num of busy-loop iterations after which 8351 * a timer interrupt should have happened in host 8352 */ 8353 #define TIMER_INTERRUPT_DELAY 100000000 8354 8355 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void) 8356 { 8357 handle_irq(0x79, irq_79_handler_guest); 8358 irq_enable(); 8359 8360 /* Signal to L1 CPU to trigger ioapic scan */ 8361 vmx_set_test_stage(1); 8362 /* Wait until L1 CPU to trigger ioapic scan */ 8363 while (vmx_get_test_stage() != 2) 8364 ; 8365 8366 /* 8367 * Wait for L0 timer interrupt to be raised while we run in L2 8368 * such that L0 will process the IOAPIC scan request before 8369 * resuming L2 8370 */ 8371 delay(TIMER_INTERRUPT_DELAY); 8372 8373 asm volatile ("int $0x79"); 8374 } 8375 8376 static void vmx_eoi_bitmap_ioapic_scan_test(void) 8377 { 8378 if (!cpu_has_apicv() || (cpu_count() < 2)) { 8379 report_skip(__func__); 8380 return; 8381 } 8382 8383 enable_vid(); 8384 8385 on_cpu_async(1, trigger_ioapic_scan_thread, NULL); 8386 test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest); 8387 8388 /* 8389 * Launch L2. 8390 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED). 8391 * In case the reason isn't VMX_VMCALL, the asserion inside 8392 * skip_exit_vmcall() will fail. 8393 */ 8394 enter_guest(); 8395 skip_exit_vmcall(); 8396 8397 /* Let L2 finish */ 8398 enter_guest(); 8399 report(1, __func__); 8400 } 8401 8402 #define HLT_WITH_RVI_VECTOR (0xf1) 8403 8404 bool vmx_hlt_with_rvi_guest_isr_fired; 8405 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs) 8406 { 8407 vmx_hlt_with_rvi_guest_isr_fired = true; 8408 eoi(); 8409 } 8410 8411 static void vmx_hlt_with_rvi_guest(void) 8412 { 8413 handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr); 8414 8415 irq_enable(); 8416 asm volatile ("nop"); 8417 8418 vmcall(); 8419 } 8420 8421 static void vmx_hlt_with_rvi_test(void) 8422 { 8423 if (!cpu_has_apicv()) { 8424 report_skip(__func__); 8425 return; 8426 } 8427 8428 enable_vid(); 8429 8430 vmx_hlt_with_rvi_guest_isr_fired = false; 8431 test_set_guest(vmx_hlt_with_rvi_guest); 8432 8433 enter_guest(); 8434 skip_exit_vmcall(); 8435 8436 vmcs_write(GUEST_ACTV_STATE, ACTV_HLT); 8437 vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR); 8438 enter_guest(); 8439 8440 report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest"); 8441 } 8442 8443 static void set_irq_line_thread(void *data) 8444 { 8445 /* Wait until other CPU entered L2 */ 8446 while (vmx_get_test_stage() != 1) 8447 ; 8448 8449 /* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */ 8450 ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL); 8451 vmx_set_test_stage(2); 8452 } 8453 8454 static bool irq_78_handler_vmcall_before_eoi; 8455 static void irq_78_handler_guest(isr_regs_t *regs) 8456 { 8457 set_irq_line(0xf, 0); 8458 if (irq_78_handler_vmcall_before_eoi) 8459 vmcall(); 8460 eoi(); 8461 vmcall(); 8462 } 8463 8464 static void vmx_apic_passthrough_guest(void) 8465 { 8466 handle_irq(0x78, irq_78_handler_guest); 8467 irq_enable(); 8468 8469 /* If requested, wait for other CPU to trigger ioapic scan */ 8470 if (vmx_get_test_stage() < 1) { 8471 vmx_set_test_stage(1); 8472 while (vmx_get_test_stage() != 2) 8473 ; 8474 } 8475 8476 set_irq_line(0xf, 1); 8477 } 8478 8479 static void vmx_apic_passthrough(bool set_irq_line_from_thread) 8480 { 8481 if (set_irq_line_from_thread && (cpu_count() < 2)) { 8482 report_skip(__func__); 8483 return; 8484 } 8485 8486 /* Test device is required for generating IRQs */ 8487 if (!test_device_enabled()) { 8488 report_skip(__func__); 8489 return; 8490 } 8491 u64 cpu_ctrl_0 = CPU_SECONDARY; 8492 u64 cpu_ctrl_1 = 0; 8493 8494 disable_intercept_for_x2apic_msrs(); 8495 8496 vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT); 8497 8498 vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0); 8499 vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1); 8500 8501 if (set_irq_line_from_thread) { 8502 irq_78_handler_vmcall_before_eoi = false; 8503 on_cpu_async(1, set_irq_line_thread, NULL); 8504 } else { 8505 irq_78_handler_vmcall_before_eoi = true; 8506 ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL); 8507 vmx_set_test_stage(2); 8508 } 8509 test_set_guest(vmx_apic_passthrough_guest); 8510 8511 if (irq_78_handler_vmcall_before_eoi) { 8512 /* Before EOI remote_irr should still be set */ 8513 enter_guest(); 8514 skip_exit_vmcall(); 8515 TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr, 8516 "IOAPIC pass-through: remote_irr=1 before EOI"); 8517 } 8518 8519 /* After EOI remote_irr should be cleared */ 8520 enter_guest(); 8521 skip_exit_vmcall(); 8522 TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr, 8523 "IOAPIC pass-through: remote_irr=0 after EOI"); 8524 8525 /* Let L2 finish */ 8526 enter_guest(); 8527 report(1, __func__); 8528 } 8529 8530 static void vmx_apic_passthrough_test(void) 8531 { 8532 vmx_apic_passthrough(false); 8533 } 8534 8535 static void vmx_apic_passthrough_thread_test(void) 8536 { 8537 vmx_apic_passthrough(true); 8538 } 8539 8540 static void vmx_apic_passthrough_tpr_threshold_guest(void) 8541 { 8542 cli(); 8543 apic_set_tpr(0); 8544 } 8545 8546 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired; 8547 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs) 8548 { 8549 vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true; 8550 eoi(); 8551 } 8552 8553 static void vmx_apic_passthrough_tpr_threshold_test(void) 8554 { 8555 int ipi_vector = 0xe1; 8556 8557 disable_intercept_for_x2apic_msrs(); 8558 vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT); 8559 8560 /* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */ 8561 cli(); 8562 apic_set_tpr((ipi_vector >> 4) + 1); 8563 apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | 8564 APIC_DM_FIXED | ipi_vector, 8565 0); 8566 8567 test_set_guest(vmx_apic_passthrough_tpr_threshold_guest); 8568 enter_guest(); 8569 8570 report(apic_get_tpr() == 0, "TPR was zero by guest"); 8571 8572 /* Clean pending self-IPI */ 8573 vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false; 8574 handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr); 8575 sti(); 8576 asm volatile ("nop"); 8577 report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired"); 8578 8579 report(1, __func__); 8580 } 8581 8582 static u64 init_signal_test_exit_reason; 8583 static bool init_signal_test_thread_continued; 8584 8585 static void init_signal_test_thread(void *data) 8586 { 8587 struct vmcs *test_vmcs = data; 8588 8589 /* Enter VMX operation (i.e. exec VMXON) */ 8590 u64 *ap_vmxon_region = alloc_page(); 8591 enable_vmx(); 8592 init_vmx(ap_vmxon_region); 8593 _vmx_on(ap_vmxon_region); 8594 8595 /* Signal CPU have entered VMX operation */ 8596 vmx_set_test_stage(1); 8597 8598 /* Wait for BSP CPU to send INIT signal */ 8599 while (vmx_get_test_stage() != 2) 8600 ; 8601 8602 /* 8603 * Signal that we continue as usual as INIT signal 8604 * should be blocked while CPU is in VMX operation 8605 */ 8606 vmx_set_test_stage(3); 8607 8608 /* Wait for signal to enter VMX non-root mode */ 8609 while (vmx_get_test_stage() != 4) 8610 ; 8611 8612 /* Enter VMX non-root mode */ 8613 test_set_guest(v2_null_test_guest); 8614 make_vmcs_current(test_vmcs); 8615 enter_guest(); 8616 /* Save exit reason for BSP CPU to compare to expected result */ 8617 init_signal_test_exit_reason = vmcs_read(EXI_REASON); 8618 /* VMCLEAR test-vmcs so it could be loaded by BSP CPU */ 8619 vmcs_clear(test_vmcs); 8620 launched = false; 8621 /* Signal that CPU exited to VMX root mode */ 8622 vmx_set_test_stage(5); 8623 8624 /* Wait for BSP CPU to signal to exit VMX operation */ 8625 while (vmx_get_test_stage() != 6) 8626 ; 8627 8628 /* Exit VMX operation (i.e. exec VMXOFF) */ 8629 vmx_off(); 8630 8631 /* 8632 * Signal to BSP CPU that we continue as usual as INIT signal 8633 * should have been consumed by VMX_INIT exit from guest 8634 */ 8635 vmx_set_test_stage(7); 8636 8637 /* Wait for BSP CPU to signal to enter VMX operation */ 8638 while (vmx_get_test_stage() != 8) 8639 ; 8640 /* Enter VMX operation (i.e. exec VMXON) */ 8641 _vmx_on(ap_vmxon_region); 8642 /* Signal to BSP we are in VMX operation */ 8643 vmx_set_test_stage(9); 8644 8645 /* Wait for BSP CPU to send INIT signal */ 8646 while (vmx_get_test_stage() != 10) 8647 ; 8648 8649 /* Exit VMX operation (i.e. exec VMXOFF) */ 8650 vmx_off(); 8651 8652 /* 8653 * Exiting VMX operation should result in latched 8654 * INIT signal being processed. Therefore, we should 8655 * never reach the below code. Thus, signal to BSP 8656 * CPU if we have reached here so it is able to 8657 * report an issue if it happens. 8658 */ 8659 init_signal_test_thread_continued = true; 8660 } 8661 8662 #define INIT_SIGNAL_TEST_DELAY 100000000ULL 8663 8664 static void vmx_init_signal_test(void) 8665 { 8666 struct vmcs *test_vmcs; 8667 8668 if (cpu_count() < 2) { 8669 report_skip(__func__); 8670 return; 8671 } 8672 8673 /* VMCLEAR test-vmcs so it could be loaded by other CPU */ 8674 vmcs_save(&test_vmcs); 8675 vmcs_clear(test_vmcs); 8676 8677 vmx_set_test_stage(0); 8678 on_cpu_async(1, init_signal_test_thread, test_vmcs); 8679 8680 /* Wait for other CPU to enter VMX operation */ 8681 while (vmx_get_test_stage() != 1) 8682 ; 8683 8684 /* Send INIT signal to other CPU */ 8685 apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT, 8686 id_map[1]); 8687 /* Signal other CPU we have sent INIT signal */ 8688 vmx_set_test_stage(2); 8689 8690 /* 8691 * Wait reasonable amount of time for INIT signal to 8692 * be received on other CPU and verify that other CPU 8693 * have proceed as usual to next test stage as INIT 8694 * signal should be blocked while other CPU in 8695 * VMX operation 8696 */ 8697 delay(INIT_SIGNAL_TEST_DELAY); 8698 report(vmx_get_test_stage() == 3, 8699 "INIT signal blocked when CPU in VMX operation"); 8700 /* No point to continue if we failed at this point */ 8701 if (vmx_get_test_stage() != 3) 8702 return; 8703 8704 /* Signal other CPU to enter VMX non-root mode */ 8705 init_signal_test_exit_reason = -1ull; 8706 vmx_set_test_stage(4); 8707 /* 8708 * Wait reasonable amont of time for other CPU 8709 * to exit to VMX root mode 8710 */ 8711 delay(INIT_SIGNAL_TEST_DELAY); 8712 if (vmx_get_test_stage() != 5) { 8713 report(false, "Pending INIT signal didn't result in VMX exit"); 8714 return; 8715 } 8716 report(init_signal_test_exit_reason == VMX_INIT, 8717 "INIT signal during VMX non-root mode result in exit-reason %s (%lu)", 8718 exit_reason_description(init_signal_test_exit_reason), 8719 init_signal_test_exit_reason); 8720 8721 /* Run guest to completion */ 8722 make_vmcs_current(test_vmcs); 8723 enter_guest(); 8724 8725 /* Signal other CPU to exit VMX operation */ 8726 init_signal_test_thread_continued = false; 8727 vmx_set_test_stage(6); 8728 8729 /* Wait reasonable amount of time for other CPU to exit VMX operation */ 8730 delay(INIT_SIGNAL_TEST_DELAY); 8731 report(vmx_get_test_stage() == 7, 8732 "INIT signal consumed on VMX_INIT exit"); 8733 /* No point to continue if we failed at this point */ 8734 if (vmx_get_test_stage() != 7) 8735 return; 8736 8737 /* Signal other CPU to enter VMX operation */ 8738 vmx_set_test_stage(8); 8739 /* Wait for other CPU to enter VMX operation */ 8740 while (vmx_get_test_stage() != 9) 8741 ; 8742 8743 /* Send INIT signal to other CPU */ 8744 apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT, 8745 id_map[1]); 8746 /* Signal other CPU we have sent INIT signal */ 8747 vmx_set_test_stage(10); 8748 8749 /* 8750 * Wait reasonable amount of time for other CPU 8751 * to exit VMX operation and process INIT signal 8752 */ 8753 delay(INIT_SIGNAL_TEST_DELAY); 8754 report(!init_signal_test_thread_continued, 8755 "INIT signal processed after exit VMX operation"); 8756 8757 /* 8758 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S) 8759 * to re-init it to kvm-unit-tests standard environment. 8760 * Somehow (?) verify that SIPI was indeed received. 8761 */ 8762 } 8763 8764 enum vmcs_access { 8765 ACCESS_VMREAD, 8766 ACCESS_VMWRITE, 8767 ACCESS_NONE, 8768 }; 8769 8770 struct vmcs_shadow_test_common { 8771 enum vmcs_access op; 8772 enum Reason reason; 8773 u64 field; 8774 u64 value; 8775 u64 flags; 8776 u64 time; 8777 } l1_l2_common; 8778 8779 static inline u64 vmread_flags(u64 field, u64 *val) 8780 { 8781 u64 flags; 8782 8783 asm volatile ("vmread %2, %1; pushf; pop %0" 8784 : "=r" (flags), "=rm" (*val) : "r" (field) : "cc"); 8785 return flags & X86_EFLAGS_ALU; 8786 } 8787 8788 static inline u64 vmwrite_flags(u64 field, u64 val) 8789 { 8790 u64 flags; 8791 8792 asm volatile ("vmwrite %1, %2; pushf; pop %0" 8793 : "=r"(flags) : "rm" (val), "r" (field) : "cc"); 8794 return flags & X86_EFLAGS_ALU; 8795 } 8796 8797 static void vmx_vmcs_shadow_test_guest(void) 8798 { 8799 struct vmcs_shadow_test_common *c = &l1_l2_common; 8800 u64 start; 8801 8802 while (c->op != ACCESS_NONE) { 8803 start = rdtsc(); 8804 switch (c->op) { 8805 default: 8806 c->flags = -1ull; 8807 break; 8808 case ACCESS_VMREAD: 8809 c->flags = vmread_flags(c->field, &c->value); 8810 break; 8811 case ACCESS_VMWRITE: 8812 c->flags = vmwrite_flags(c->field, 0); 8813 break; 8814 } 8815 c->time = rdtsc() - start; 8816 vmcall(); 8817 } 8818 } 8819 8820 static u64 vmread_from_shadow(u64 field) 8821 { 8822 struct vmcs *primary; 8823 struct vmcs *shadow; 8824 u64 value; 8825 8826 TEST_ASSERT(!vmcs_save(&primary)); 8827 shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR); 8828 TEST_ASSERT(!make_vmcs_current(shadow)); 8829 value = vmcs_read(field); 8830 TEST_ASSERT(!make_vmcs_current(primary)); 8831 return value; 8832 } 8833 8834 static u64 vmwrite_to_shadow(u64 field, u64 value) 8835 { 8836 struct vmcs *primary; 8837 struct vmcs *shadow; 8838 8839 TEST_ASSERT(!vmcs_save(&primary)); 8840 shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR); 8841 TEST_ASSERT(!make_vmcs_current(shadow)); 8842 vmcs_write(field, value); 8843 value = vmcs_read(field); 8844 TEST_ASSERT(!make_vmcs_current(primary)); 8845 return value; 8846 } 8847 8848 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access) 8849 { 8850 struct vmcs_shadow_test_common *c = &l1_l2_common; 8851 8852 c->op = access; 8853 vmcs_write(VMX_INST_ERROR, 0); 8854 enter_guest(); 8855 c->reason = vmcs_read(EXI_REASON) & 0xffff; 8856 if (c->reason != VMX_VMCALL) { 8857 skip_exit_insn(); 8858 enter_guest(); 8859 } 8860 skip_exit_vmcall(); 8861 } 8862 8863 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field) 8864 { 8865 struct vmcs_shadow_test_common *c = &l1_l2_common; 8866 struct vmcs *shadow; 8867 u64 value; 8868 uintptr_t flags[2]; 8869 bool good_shadow; 8870 u32 vmx_inst_error; 8871 8872 report_prefix_pushf("field %lx", field); 8873 c->field = field; 8874 8875 shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR); 8876 if (shadow != (struct vmcs *)-1ull) { 8877 flags[ACCESS_VMREAD] = vmread_flags(field, &value); 8878 flags[ACCESS_VMWRITE] = vmwrite_flags(field, value); 8879 good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE]; 8880 } else { 8881 /* 8882 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on 8883 * shadowed-fields should fail with setting RFLAGS.CF. 8884 */ 8885 flags[ACCESS_VMREAD] = X86_EFLAGS_CF; 8886 flags[ACCESS_VMWRITE] = X86_EFLAGS_CF; 8887 good_shadow = false; 8888 } 8889 8890 /* Intercept both VMREAD and VMWRITE. */ 8891 report_prefix_push("no VMREAD/VMWRITE permission"); 8892 /* VMWRITE/VMREAD done on reserved-bit should always intercept */ 8893 if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) { 8894 set_bit(field, bitmap[ACCESS_VMREAD]); 8895 set_bit(field, bitmap[ACCESS_VMWRITE]); 8896 } 8897 vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE); 8898 report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE"); 8899 vmcs_shadow_test_access(bitmap, ACCESS_VMREAD); 8900 report(c->reason == VMX_VMREAD, "not shadowed for VMREAD"); 8901 report_prefix_pop(); 8902 8903 if (field >> VMCS_FIELD_RESERVED_SHIFT) 8904 goto out; 8905 8906 /* Permit shadowed VMREAD. */ 8907 report_prefix_push("VMREAD permission only"); 8908 clear_bit(field, bitmap[ACCESS_VMREAD]); 8909 set_bit(field, bitmap[ACCESS_VMWRITE]); 8910 if (good_shadow) 8911 value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field); 8912 vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE); 8913 report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE"); 8914 vmcs_shadow_test_access(bitmap, ACCESS_VMREAD); 8915 vmx_inst_error = vmcs_read(VMX_INST_ERROR); 8916 report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)", 8917 c->time); 8918 report(c->flags == flags[ACCESS_VMREAD], 8919 "ALU flags after VMREAD (%lx) are as expected (%lx)", 8920 c->flags, flags[ACCESS_VMREAD]); 8921 if (good_shadow) 8922 report(c->value == value, 8923 "value read from shadow (%lx) is as expected (%lx)", 8924 c->value, value); 8925 else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD]) 8926 report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT, 8927 "VMX_INST_ERROR (%d) is as expected (%d)", 8928 vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT); 8929 report_prefix_pop(); 8930 8931 /* Permit shadowed VMWRITE. */ 8932 report_prefix_push("VMWRITE permission only"); 8933 set_bit(field, bitmap[ACCESS_VMREAD]); 8934 clear_bit(field, bitmap[ACCESS_VMWRITE]); 8935 if (good_shadow) 8936 vmwrite_to_shadow(field, MAGIC_VAL_1 + field); 8937 vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE); 8938 vmx_inst_error = vmcs_read(VMX_INST_ERROR); 8939 report(c->reason == VMX_VMCALL, 8940 "shadowed for VMWRITE (in %ld cycles)", 8941 c->time); 8942 report(c->flags == flags[ACCESS_VMREAD], 8943 "ALU flags after VMWRITE (%lx) are as expected (%lx)", 8944 c->flags, flags[ACCESS_VMREAD]); 8945 if (good_shadow) { 8946 value = vmread_from_shadow(field); 8947 report(value == 0, 8948 "shadow VMCS value (%lx) is as expected (%lx)", value, 8949 0ul); 8950 } else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) { 8951 report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT, 8952 "VMX_INST_ERROR (%d) is as expected (%d)", 8953 vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT); 8954 } 8955 vmcs_shadow_test_access(bitmap, ACCESS_VMREAD); 8956 report(c->reason == VMX_VMREAD, "not shadowed for VMREAD"); 8957 report_prefix_pop(); 8958 8959 /* Permit shadowed VMREAD and VMWRITE. */ 8960 report_prefix_push("VMREAD and VMWRITE permission"); 8961 clear_bit(field, bitmap[ACCESS_VMREAD]); 8962 clear_bit(field, bitmap[ACCESS_VMWRITE]); 8963 if (good_shadow) 8964 vmwrite_to_shadow(field, MAGIC_VAL_1 + field); 8965 vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE); 8966 vmx_inst_error = vmcs_read(VMX_INST_ERROR); 8967 report(c->reason == VMX_VMCALL, 8968 "shadowed for VMWRITE (in %ld cycles)", 8969 c->time); 8970 report(c->flags == flags[ACCESS_VMREAD], 8971 "ALU flags after VMWRITE (%lx) are as expected (%lx)", 8972 c->flags, flags[ACCESS_VMREAD]); 8973 if (good_shadow) { 8974 value = vmread_from_shadow(field); 8975 report(value == 0, 8976 "shadow VMCS value (%lx) is as expected (%lx)", value, 8977 0ul); 8978 } else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) { 8979 report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT, 8980 "VMX_INST_ERROR (%d) is as expected (%d)", 8981 vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT); 8982 } 8983 vmcs_shadow_test_access(bitmap, ACCESS_VMREAD); 8984 vmx_inst_error = vmcs_read(VMX_INST_ERROR); 8985 report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)", 8986 c->time); 8987 report(c->flags == flags[ACCESS_VMREAD], 8988 "ALU flags after VMREAD (%lx) are as expected (%lx)", 8989 c->flags, flags[ACCESS_VMREAD]); 8990 if (good_shadow) 8991 report(c->value == 0, 8992 "value read from shadow (%lx) is as expected (%lx)", 8993 c->value, 0ul); 8994 else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD]) 8995 report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT, 8996 "VMX_INST_ERROR (%d) is as expected (%d)", 8997 vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT); 8998 report_prefix_pop(); 8999 9000 out: 9001 report_prefix_pop(); 9002 } 9003 9004 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2]) 9005 { 9006 unsigned base; 9007 unsigned index; 9008 unsigned bit; 9009 unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM); 9010 9011 /* Run test on all possible valid VMCS fields */ 9012 for (base = 0; 9013 base < (1 << VMCS_FIELD_RESERVED_SHIFT); 9014 base += (1 << VMCS_FIELD_TYPE_SHIFT)) 9015 for (index = 0; index <= highest_index; index++) 9016 vmcs_shadow_test_field(bitmap, base + index); 9017 9018 /* 9019 * Run tests on some invalid VMCS fields 9020 * (Have reserved bit set). 9021 */ 9022 for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++) 9023 vmcs_shadow_test_field(bitmap, (1ull << bit)); 9024 } 9025 9026 static void vmx_vmcs_shadow_test(void) 9027 { 9028 u8 *bitmap[2]; 9029 struct vmcs *shadow; 9030 9031 if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) { 9032 printf("\t'Activate secondary controls' not supported.\n"); 9033 return; 9034 } 9035 9036 if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) { 9037 printf("\t'VMCS shadowing' not supported.\n"); 9038 return; 9039 } 9040 9041 if (!(rdmsr(MSR_IA32_VMX_MISC) & 9042 MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) { 9043 printf("\tVMWRITE can't modify VM-exit information fields.\n"); 9044 return; 9045 } 9046 9047 test_set_guest(vmx_vmcs_shadow_test_guest); 9048 9049 bitmap[ACCESS_VMREAD] = alloc_page(); 9050 bitmap[ACCESS_VMWRITE] = alloc_page(); 9051 9052 vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD])); 9053 vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE])); 9054 9055 shadow = alloc_page(); 9056 shadow->hdr.revision_id = basic.revision; 9057 shadow->hdr.shadow_vmcs = 1; 9058 TEST_ASSERT(!vmcs_clear(shadow)); 9059 9060 vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC); 9061 vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY); 9062 vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS); 9063 9064 vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow)); 9065 report_prefix_push("valid link pointer"); 9066 vmx_vmcs_shadow_test_body(bitmap); 9067 report_prefix_pop(); 9068 9069 vmcs_write(VMCS_LINK_PTR, -1ull); 9070 report_prefix_push("invalid link pointer"); 9071 vmx_vmcs_shadow_test_body(bitmap); 9072 report_prefix_pop(); 9073 9074 l1_l2_common.op = ACCESS_NONE; 9075 enter_guest(); 9076 } 9077 9078 /* 9079 * This test monitors the difference between a guest RDTSC instruction 9080 * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12 9081 * VM-exit MSR-store list when taking a VM-exit on the instruction 9082 * following RDTSC. 9083 */ 9084 #define RDTSC_DIFF_ITERS 100000 9085 #define RDTSC_DIFF_FAILS 100 9086 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750 9087 9088 /* 9089 * Set 'use TSC offsetting' and set the guest offset to the 9090 * inverse of the host's current TSC value, so that the guest starts running 9091 * with an effective TSC value of 0. 9092 */ 9093 static void reset_guest_tsc_to_zero(void) 9094 { 9095 TEST_ASSERT_MSG(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET, 9096 "Expected support for 'use TSC offsetting'"); 9097 9098 vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET); 9099 vmcs_write(TSC_OFFSET, -rdtsc()); 9100 } 9101 9102 static void rdtsc_vmexit_diff_test_guest(void) 9103 { 9104 int i; 9105 9106 for (i = 0; i < RDTSC_DIFF_ITERS; i++) 9107 /* Ensure rdtsc is the last instruction before the vmcall. */ 9108 asm volatile("rdtsc; vmcall" : : : "eax", "edx"); 9109 } 9110 9111 /* 9112 * This function only considers the "use TSC offsetting" VM-execution 9113 * control. It does not handle "use TSC scaling" (because the latter 9114 * isn't available to the host today.) 9115 */ 9116 static unsigned long long host_time_to_guest_time(unsigned long long t) 9117 { 9118 TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) || 9119 !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING)); 9120 9121 if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET) 9122 t += vmcs_read(TSC_OFFSET); 9123 9124 return t; 9125 } 9126 9127 static unsigned long long rdtsc_vmexit_diff_test_iteration(void) 9128 { 9129 unsigned long long guest_tsc, host_to_guest_tsc; 9130 9131 enter_guest(); 9132 skip_exit_vmcall(); 9133 guest_tsc = (u32) regs.rax + (regs.rdx << 32); 9134 host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value); 9135 9136 return host_to_guest_tsc - guest_tsc; 9137 } 9138 9139 static void rdtsc_vmexit_diff_test(void) 9140 { 9141 int fail = 0; 9142 int i; 9143 9144 test_set_guest(rdtsc_vmexit_diff_test_guest); 9145 9146 reset_guest_tsc_to_zero(); 9147 9148 /* 9149 * Set up the VMCS12 VM-exit MSR-store list to store just one 9150 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is 9151 * in the host time domain (i.e., it is not adjusted according 9152 * to the TSC multiplier and TSC offset fields in the VMCS12, 9153 * as a guest RDTSC would be.) 9154 */ 9155 exit_msr_store = alloc_page(); 9156 exit_msr_store[0].index = MSR_IA32_TSC; 9157 vmcs_write(EXI_MSR_ST_CNT, 1); 9158 vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store)); 9159 9160 for (i = 0; i < RDTSC_DIFF_ITERS; i++) { 9161 if (rdtsc_vmexit_diff_test_iteration() >= 9162 HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD) 9163 fail++; 9164 } 9165 9166 enter_guest(); 9167 9168 report(fail < RDTSC_DIFF_FAILS, 9169 "RDTSC to VM-exit delta too high in %d of %d iterations", 9170 fail, RDTSC_DIFF_ITERS); 9171 } 9172 9173 static int invalid_msr_init(struct vmcs *vmcs) 9174 { 9175 if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) { 9176 printf("\tPreemption timer is not supported\n"); 9177 return VMX_TEST_EXIT; 9178 } 9179 vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT); 9180 preempt_val = 10000000; 9181 vmcs_write(PREEMPT_TIMER_VALUE, preempt_val); 9182 preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F; 9183 9184 if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT)) 9185 printf("\tSave preemption value is not supported\n"); 9186 9187 vmcs_write(ENT_MSR_LD_CNT, 1); 9188 vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000); 9189 9190 return VMX_TEST_START; 9191 } 9192 9193 9194 static void invalid_msr_main(void) 9195 { 9196 report(0, "Invalid MSR load"); 9197 } 9198 9199 static int invalid_msr_exit_handler(void) 9200 { 9201 report(0, "Invalid MSR load"); 9202 print_vmexit_info(); 9203 return VMX_TEST_EXIT; 9204 } 9205 9206 static int invalid_msr_entry_failure(struct vmentry_failure *failure) 9207 { 9208 ulong reason; 9209 9210 reason = vmcs_read(EXI_REASON); 9211 report(reason == (0x80000000u | VMX_FAIL_MSR), "Invalid MSR load"); 9212 return VMX_TEST_VMEXIT; 9213 } 9214 9215 /* 9216 * The max number of MSRs in an atomic switch MSR list is: 9217 * (111B + 1) * 512 = 4096 9218 * 9219 * Each list entry consumes: 9220 * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes 9221 * 9222 * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some. 9223 */ 9224 static const u32 msr_list_page_order = 5; 9225 9226 static void atomic_switch_msr_limit_test_guest(void) 9227 { 9228 vmcall(); 9229 } 9230 9231 static void populate_msr_list(struct vmx_msr_entry *msr_list, 9232 size_t byte_capacity, int count) 9233 { 9234 int i; 9235 9236 for (i = 0; i < count; i++) { 9237 msr_list[i].index = MSR_IA32_TSC; 9238 msr_list[i].reserved = 0; 9239 msr_list[i].value = 0x1234567890abcdef; 9240 } 9241 9242 memset(msr_list + count, 0xff, 9243 byte_capacity - count * sizeof(*msr_list)); 9244 } 9245 9246 static int max_msr_list_size(void) 9247 { 9248 u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC); 9249 u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1; 9250 9251 return factor * 512; 9252 } 9253 9254 static void atomic_switch_msrs_test(int count) 9255 { 9256 struct vmx_msr_entry *vm_enter_load; 9257 struct vmx_msr_entry *vm_exit_load; 9258 struct vmx_msr_entry *vm_exit_store; 9259 int max_allowed = max_msr_list_size(); 9260 int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT); 9261 /* Exceeding the max MSR list size at exit trigers KVM to abort. */ 9262 int exit_count = count > max_allowed ? max_allowed : count; 9263 int cleanup_count = count > max_allowed ? 2 : 1; 9264 int i; 9265 9266 /* 9267 * Check for the IA32_TSC MSR, 9268 * available with the "TSC flag" and used to populate the MSR lists. 9269 */ 9270 if (!(cpuid(1).d & (1 << 4))) { 9271 report_skip(__func__); 9272 return; 9273 } 9274 9275 /* Set L2 guest. */ 9276 test_set_guest(atomic_switch_msr_limit_test_guest); 9277 9278 /* Setup atomic MSR switch lists. */ 9279 vm_enter_load = alloc_pages(msr_list_page_order); 9280 vm_exit_load = alloc_pages(msr_list_page_order); 9281 vm_exit_store = alloc_pages(msr_list_page_order); 9282 9283 vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load); 9284 vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load); 9285 vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store); 9286 9287 /* 9288 * VM-Enter should succeed up to the max number of MSRs per list, and 9289 * should not consume junk beyond the last entry. 9290 */ 9291 populate_msr_list(vm_enter_load, byte_capacity, count); 9292 populate_msr_list(vm_exit_load, byte_capacity, exit_count); 9293 populate_msr_list(vm_exit_store, byte_capacity, exit_count); 9294 9295 vmcs_write(ENT_MSR_LD_CNT, count); 9296 vmcs_write(EXI_MSR_LD_CNT, exit_count); 9297 vmcs_write(EXI_MSR_ST_CNT, exit_count); 9298 9299 if (count <= max_allowed) { 9300 enter_guest(); 9301 assert_exit_reason(VMX_VMCALL); 9302 skip_exit_vmcall(); 9303 } else { 9304 u32 exit_reason; 9305 u32 exit_reason_want; 9306 u32 exit_qual; 9307 9308 enter_guest_with_invalid_guest_state(); 9309 9310 exit_reason = vmcs_read(EXI_REASON); 9311 exit_reason_want = VMX_FAIL_MSR | VMX_ENTRY_FAILURE; 9312 report(exit_reason == exit_reason_want, 9313 "exit_reason, %u, is %u.", exit_reason, 9314 exit_reason_want); 9315 9316 exit_qual = vmcs_read(EXI_QUALIFICATION); 9317 report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.", 9318 exit_qual, max_allowed + 1); 9319 } 9320 9321 /* Cleanup. */ 9322 vmcs_write(ENT_MSR_LD_CNT, 0); 9323 vmcs_write(EXI_MSR_LD_CNT, 0); 9324 vmcs_write(EXI_MSR_ST_CNT, 0); 9325 for (i = 0; i < cleanup_count; i++) { 9326 enter_guest(); 9327 skip_exit_vmcall(); 9328 } 9329 free_pages_by_order(vm_enter_load, msr_list_page_order); 9330 free_pages_by_order(vm_exit_load, msr_list_page_order); 9331 free_pages_by_order(vm_exit_store, msr_list_page_order); 9332 } 9333 9334 static void atomic_switch_max_msrs_test(void) 9335 { 9336 atomic_switch_msrs_test(max_msr_list_size()); 9337 } 9338 9339 static void atomic_switch_overflow_msrs_test(void) 9340 { 9341 atomic_switch_msrs_test(max_msr_list_size() + 1); 9342 } 9343 9344 #define TEST(name) { #name, .v2 = name } 9345 9346 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */ 9347 struct vmx_test vmx_tests[] = { 9348 { "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} }, 9349 { "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} }, 9350 { "preemption timer", preemption_timer_init, preemption_timer_main, 9351 preemption_timer_exit_handler, NULL, {0} }, 9352 { "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main, 9353 test_ctrl_pat_exit_handler, NULL, {0} }, 9354 { "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main, 9355 test_ctrl_efer_exit_handler, NULL, {0} }, 9356 { "CR shadowing", NULL, cr_shadowing_main, 9357 cr_shadowing_exit_handler, NULL, {0} }, 9358 { "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler, 9359 NULL, {0} }, 9360 { "instruction intercept", insn_intercept_init, insn_intercept_main, 9361 insn_intercept_exit_handler, NULL, {0} }, 9362 { "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} }, 9363 { "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} }, 9364 { "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} }, 9365 { "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} }, 9366 { "interrupt", interrupt_init, interrupt_main, 9367 interrupt_exit_handler, NULL, {0} }, 9368 { "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler, 9369 NULL, {0} }, 9370 { "MSR switch", msr_switch_init, msr_switch_main, 9371 msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure }, 9372 { "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} }, 9373 { "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main, 9374 disable_rdtscp_exit_handler, NULL, {0} }, 9375 { "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} }, 9376 { "into", into_init, into_guest_main, into_exit_handler, NULL, {0} }, 9377 { "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main, 9378 exit_monitor_from_l2_handler, NULL, {0} }, 9379 { "invalid_msr", invalid_msr_init, invalid_msr_main, 9380 invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure}, 9381 /* Basic V2 tests. */ 9382 TEST(v2_null_test), 9383 TEST(v2_multiple_entries_test), 9384 TEST(fixture_test_case1), 9385 TEST(fixture_test_case2), 9386 /* Opcode tests. */ 9387 TEST(invvpid_test_v2), 9388 /* VM-entry tests */ 9389 TEST(vmx_controls_test), 9390 TEST(vmx_host_state_area_test), 9391 TEST(vmx_guest_state_area_test), 9392 TEST(vmentry_movss_shadow_test), 9393 /* APICv tests */ 9394 TEST(vmx_eoi_bitmap_ioapic_scan_test), 9395 TEST(vmx_hlt_with_rvi_test), 9396 TEST(apic_reg_virt_test), 9397 TEST(virt_x2apic_mode_test), 9398 /* APIC pass-through tests */ 9399 TEST(vmx_apic_passthrough_test), 9400 TEST(vmx_apic_passthrough_thread_test), 9401 TEST(vmx_apic_passthrough_tpr_threshold_test), 9402 TEST(vmx_init_signal_test), 9403 /* VMCS Shadowing tests */ 9404 TEST(vmx_vmcs_shadow_test), 9405 /* Regression tests */ 9406 TEST(vmx_cr_load_test), 9407 TEST(vmx_nm_test), 9408 TEST(vmx_db_test), 9409 TEST(vmx_nmi_window_test), 9410 TEST(vmx_intr_window_test), 9411 TEST(vmx_pending_event_test), 9412 TEST(vmx_pending_event_hlt_test), 9413 TEST(vmx_store_tsc_test), 9414 /* EPT access tests. */ 9415 TEST(ept_access_test_not_present), 9416 TEST(ept_access_test_read_only), 9417 TEST(ept_access_test_write_only), 9418 TEST(ept_access_test_read_write), 9419 TEST(ept_access_test_execute_only), 9420 TEST(ept_access_test_read_execute), 9421 TEST(ept_access_test_write_execute), 9422 TEST(ept_access_test_read_write_execute), 9423 TEST(ept_access_test_reserved_bits), 9424 TEST(ept_access_test_ignored_bits), 9425 TEST(ept_access_test_paddr_not_present_ad_disabled), 9426 TEST(ept_access_test_paddr_not_present_ad_enabled), 9427 TEST(ept_access_test_paddr_read_only_ad_disabled), 9428 TEST(ept_access_test_paddr_read_only_ad_enabled), 9429 TEST(ept_access_test_paddr_read_write), 9430 TEST(ept_access_test_paddr_read_write_execute), 9431 TEST(ept_access_test_paddr_read_execute_ad_disabled), 9432 TEST(ept_access_test_paddr_read_execute_ad_enabled), 9433 TEST(ept_access_test_paddr_not_present_page_fault), 9434 TEST(ept_access_test_force_2m_page), 9435 /* Atomic MSR switch tests. */ 9436 TEST(atomic_switch_max_msrs_test), 9437 TEST(atomic_switch_overflow_msrs_test), 9438 TEST(rdtsc_vmexit_diff_test), 9439 { NULL, NULL, NULL, NULL, NULL, {0} }, 9440 }; 9441