1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * tools/testing/selftests/kvm/lib/x86_64/processor.c 4 * 5 * Copyright (C) 2018, Google LLC. 6 */ 7 8 #include "test_util.h" 9 #include "kvm_util.h" 10 #include "../kvm_util_internal.h" 11 #include "processor.h" 12 13 #ifndef NUM_INTERRUPTS 14 #define NUM_INTERRUPTS 256 15 #endif 16 17 #define DEFAULT_CODE_SELECTOR 0x8 18 #define DEFAULT_DATA_SELECTOR 0x10 19 20 vm_vaddr_t exception_handlers; 21 22 /* Virtual translation table structure declarations */ 23 struct pageUpperEntry { 24 uint64_t present:1; 25 uint64_t writable:1; 26 uint64_t user:1; 27 uint64_t write_through:1; 28 uint64_t cache_disable:1; 29 uint64_t accessed:1; 30 uint64_t ignored_06:1; 31 uint64_t page_size:1; 32 uint64_t ignored_11_08:4; 33 uint64_t pfn:40; 34 uint64_t ignored_62_52:11; 35 uint64_t execute_disable:1; 36 }; 37 38 struct pageTableEntry { 39 uint64_t present:1; 40 uint64_t writable:1; 41 uint64_t user:1; 42 uint64_t write_through:1; 43 uint64_t cache_disable:1; 44 uint64_t accessed:1; 45 uint64_t dirty:1; 46 uint64_t reserved_07:1; 47 uint64_t global:1; 48 uint64_t ignored_11_09:3; 49 uint64_t pfn:40; 50 uint64_t ignored_62_52:11; 51 uint64_t execute_disable:1; 52 }; 53 54 void regs_dump(FILE *stream, struct kvm_regs *regs, 55 uint8_t indent) 56 { 57 fprintf(stream, "%*srax: 0x%.16llx rbx: 0x%.16llx " 58 "rcx: 0x%.16llx rdx: 0x%.16llx\n", 59 indent, "", 60 regs->rax, regs->rbx, regs->rcx, regs->rdx); 61 fprintf(stream, "%*srsi: 0x%.16llx rdi: 0x%.16llx " 62 "rsp: 0x%.16llx rbp: 0x%.16llx\n", 63 indent, "", 64 regs->rsi, regs->rdi, regs->rsp, regs->rbp); 65 fprintf(stream, "%*sr8: 0x%.16llx r9: 0x%.16llx " 66 "r10: 0x%.16llx r11: 0x%.16llx\n", 67 indent, "", 68 regs->r8, regs->r9, regs->r10, regs->r11); 69 fprintf(stream, "%*sr12: 0x%.16llx r13: 0x%.16llx " 70 "r14: 0x%.16llx r15: 0x%.16llx\n", 71 indent, "", 72 regs->r12, regs->r13, regs->r14, regs->r15); 73 fprintf(stream, "%*srip: 0x%.16llx rfl: 0x%.16llx\n", 74 indent, "", 75 regs->rip, regs->rflags); 76 } 77 78 /* 79 * Segment Dump 80 * 81 * Input Args: 82 * stream - Output FILE stream 83 * segment - KVM segment 84 * indent - Left margin indent amount 85 * 86 * Output Args: None 87 * 88 * Return: None 89 * 90 * Dumps the state of the KVM segment given by @segment, to the FILE stream 91 * given by @stream. 92 */ 93 static void segment_dump(FILE *stream, struct kvm_segment *segment, 94 uint8_t indent) 95 { 96 fprintf(stream, "%*sbase: 0x%.16llx limit: 0x%.8x " 97 "selector: 0x%.4x type: 0x%.2x\n", 98 indent, "", segment->base, segment->limit, 99 segment->selector, segment->type); 100 fprintf(stream, "%*spresent: 0x%.2x dpl: 0x%.2x " 101 "db: 0x%.2x s: 0x%.2x l: 0x%.2x\n", 102 indent, "", segment->present, segment->dpl, 103 segment->db, segment->s, segment->l); 104 fprintf(stream, "%*sg: 0x%.2x avl: 0x%.2x " 105 "unusable: 0x%.2x padding: 0x%.2x\n", 106 indent, "", segment->g, segment->avl, 107 segment->unusable, segment->padding); 108 } 109 110 /* 111 * dtable Dump 112 * 113 * Input Args: 114 * stream - Output FILE stream 115 * dtable - KVM dtable 116 * indent - Left margin indent amount 117 * 118 * Output Args: None 119 * 120 * Return: None 121 * 122 * Dumps the state of the KVM dtable given by @dtable, to the FILE stream 123 * given by @stream. 124 */ 125 static void dtable_dump(FILE *stream, struct kvm_dtable *dtable, 126 uint8_t indent) 127 { 128 fprintf(stream, "%*sbase: 0x%.16llx limit: 0x%.4x " 129 "padding: 0x%.4x 0x%.4x 0x%.4x\n", 130 indent, "", dtable->base, dtable->limit, 131 dtable->padding[0], dtable->padding[1], dtable->padding[2]); 132 } 133 134 void sregs_dump(FILE *stream, struct kvm_sregs *sregs, 135 uint8_t indent) 136 { 137 unsigned int i; 138 139 fprintf(stream, "%*scs:\n", indent, ""); 140 segment_dump(stream, &sregs->cs, indent + 2); 141 fprintf(stream, "%*sds:\n", indent, ""); 142 segment_dump(stream, &sregs->ds, indent + 2); 143 fprintf(stream, "%*ses:\n", indent, ""); 144 segment_dump(stream, &sregs->es, indent + 2); 145 fprintf(stream, "%*sfs:\n", indent, ""); 146 segment_dump(stream, &sregs->fs, indent + 2); 147 fprintf(stream, "%*sgs:\n", indent, ""); 148 segment_dump(stream, &sregs->gs, indent + 2); 149 fprintf(stream, "%*sss:\n", indent, ""); 150 segment_dump(stream, &sregs->ss, indent + 2); 151 fprintf(stream, "%*str:\n", indent, ""); 152 segment_dump(stream, &sregs->tr, indent + 2); 153 fprintf(stream, "%*sldt:\n", indent, ""); 154 segment_dump(stream, &sregs->ldt, indent + 2); 155 156 fprintf(stream, "%*sgdt:\n", indent, ""); 157 dtable_dump(stream, &sregs->gdt, indent + 2); 158 fprintf(stream, "%*sidt:\n", indent, ""); 159 dtable_dump(stream, &sregs->idt, indent + 2); 160 161 fprintf(stream, "%*scr0: 0x%.16llx cr2: 0x%.16llx " 162 "cr3: 0x%.16llx cr4: 0x%.16llx\n", 163 indent, "", 164 sregs->cr0, sregs->cr2, sregs->cr3, sregs->cr4); 165 fprintf(stream, "%*scr8: 0x%.16llx efer: 0x%.16llx " 166 "apic_base: 0x%.16llx\n", 167 indent, "", 168 sregs->cr8, sregs->efer, sregs->apic_base); 169 170 fprintf(stream, "%*sinterrupt_bitmap:\n", indent, ""); 171 for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++) { 172 fprintf(stream, "%*s%.16llx\n", indent + 2, "", 173 sregs->interrupt_bitmap[i]); 174 } 175 } 176 177 void virt_pgd_alloc(struct kvm_vm *vm) 178 { 179 TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, "Attempt to use " 180 "unknown or unsupported guest mode, mode: 0x%x", vm->mode); 181 182 /* If needed, create page map l4 table. */ 183 if (!vm->pgd_created) { 184 vm->pgd = vm_alloc_page_table(vm); 185 vm->pgd_created = true; 186 } 187 } 188 189 static void *virt_get_pte(struct kvm_vm *vm, uint64_t pt_pfn, uint64_t vaddr, 190 int level) 191 { 192 uint64_t *page_table = addr_gpa2hva(vm, pt_pfn << vm->page_shift); 193 int index = vaddr >> (vm->page_shift + level * 9) & 0x1ffu; 194 195 return &page_table[index]; 196 } 197 198 static struct pageUpperEntry *virt_create_upper_pte(struct kvm_vm *vm, 199 uint64_t pt_pfn, 200 uint64_t vaddr, 201 uint64_t paddr, 202 int level, 203 enum x86_page_size page_size) 204 { 205 struct pageUpperEntry *pte = virt_get_pte(vm, pt_pfn, vaddr, level); 206 207 if (!pte->present) { 208 pte->writable = true; 209 pte->present = true; 210 pte->page_size = (level == page_size); 211 if (pte->page_size) 212 pte->pfn = paddr >> vm->page_shift; 213 else 214 pte->pfn = vm_alloc_page_table(vm) >> vm->page_shift; 215 } else { 216 /* 217 * Entry already present. Assert that the caller doesn't want 218 * a hugepage at this level, and that there isn't a hugepage at 219 * this level. 220 */ 221 TEST_ASSERT(level != page_size, 222 "Cannot create hugepage at level: %u, vaddr: 0x%lx\n", 223 page_size, vaddr); 224 TEST_ASSERT(!pte->page_size, 225 "Cannot create page table at level: %u, vaddr: 0x%lx\n", 226 level, vaddr); 227 } 228 return pte; 229 } 230 231 void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, 232 enum x86_page_size page_size) 233 { 234 const uint64_t pg_size = 1ull << ((page_size * 9) + 12); 235 struct pageUpperEntry *pml4e, *pdpe, *pde; 236 struct pageTableEntry *pte; 237 238 TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, 239 "Unknown or unsupported guest mode, mode: 0x%x", vm->mode); 240 241 TEST_ASSERT((vaddr % pg_size) == 0, 242 "Virtual address not aligned,\n" 243 "vaddr: 0x%lx page size: 0x%lx", vaddr, pg_size); 244 TEST_ASSERT(sparsebit_is_set(vm->vpages_valid, (vaddr >> vm->page_shift)), 245 "Invalid virtual address, vaddr: 0x%lx", vaddr); 246 TEST_ASSERT((paddr % pg_size) == 0, 247 "Physical address not aligned,\n" 248 " paddr: 0x%lx page size: 0x%lx", paddr, pg_size); 249 TEST_ASSERT((paddr >> vm->page_shift) <= vm->max_gfn, 250 "Physical address beyond maximum supported,\n" 251 " paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x", 252 paddr, vm->max_gfn, vm->page_size); 253 254 /* 255 * Allocate upper level page tables, if not already present. Return 256 * early if a hugepage was created. 257 */ 258 pml4e = virt_create_upper_pte(vm, vm->pgd >> vm->page_shift, 259 vaddr, paddr, 3, page_size); 260 if (pml4e->page_size) 261 return; 262 263 pdpe = virt_create_upper_pte(vm, pml4e->pfn, vaddr, paddr, 2, page_size); 264 if (pdpe->page_size) 265 return; 266 267 pde = virt_create_upper_pte(vm, pdpe->pfn, vaddr, paddr, 1, page_size); 268 if (pde->page_size) 269 return; 270 271 /* Fill in page table entry. */ 272 pte = virt_get_pte(vm, pde->pfn, vaddr, 0); 273 TEST_ASSERT(!pte->present, 274 "PTE already present for 4k page at vaddr: 0x%lx\n", vaddr); 275 pte->pfn = paddr >> vm->page_shift; 276 pte->writable = true; 277 pte->present = 1; 278 } 279 280 void virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr) 281 { 282 __virt_pg_map(vm, vaddr, paddr, X86_PAGE_SIZE_4K); 283 } 284 285 static struct pageTableEntry *_vm_get_page_table_entry(struct kvm_vm *vm, int vcpuid, 286 uint64_t vaddr) 287 { 288 uint16_t index[4]; 289 struct pageUpperEntry *pml4e, *pdpe, *pde; 290 struct pageTableEntry *pte; 291 struct kvm_cpuid_entry2 *entry; 292 struct kvm_sregs sregs; 293 int max_phy_addr; 294 /* Set the bottom 52 bits. */ 295 uint64_t rsvd_mask = 0x000fffffffffffff; 296 297 entry = kvm_get_supported_cpuid_index(0x80000008, 0); 298 max_phy_addr = entry->eax & 0x000000ff; 299 /* Clear the bottom bits of the reserved mask. */ 300 rsvd_mask = (rsvd_mask >> max_phy_addr) << max_phy_addr; 301 302 /* 303 * SDM vol 3, fig 4-11 "Formats of CR3 and Paging-Structure Entries 304 * with 4-Level Paging and 5-Level Paging". 305 * If IA32_EFER.NXE = 0 and the P flag of a paging-structure entry is 1, 306 * the XD flag (bit 63) is reserved. 307 */ 308 vcpu_sregs_get(vm, vcpuid, &sregs); 309 if ((sregs.efer & EFER_NX) == 0) { 310 rsvd_mask |= (1ull << 63); 311 } 312 313 TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, "Attempt to use " 314 "unknown or unsupported guest mode, mode: 0x%x", vm->mode); 315 TEST_ASSERT(sparsebit_is_set(vm->vpages_valid, 316 (vaddr >> vm->page_shift)), 317 "Invalid virtual address, vaddr: 0x%lx", 318 vaddr); 319 /* 320 * Based on the mode check above there are 48 bits in the vaddr, so 321 * shift 16 to sign extend the last bit (bit-47), 322 */ 323 TEST_ASSERT(vaddr == (((int64_t)vaddr << 16) >> 16), 324 "Canonical check failed. The virtual address is invalid."); 325 326 index[0] = (vaddr >> 12) & 0x1ffu; 327 index[1] = (vaddr >> 21) & 0x1ffu; 328 index[2] = (vaddr >> 30) & 0x1ffu; 329 index[3] = (vaddr >> 39) & 0x1ffu; 330 331 pml4e = addr_gpa2hva(vm, vm->pgd); 332 TEST_ASSERT(pml4e[index[3]].present, 333 "Expected pml4e to be present for gva: 0x%08lx", vaddr); 334 TEST_ASSERT((*(uint64_t*)(&pml4e[index[3]]) & 335 (rsvd_mask | (1ull << 7))) == 0, 336 "Unexpected reserved bits set."); 337 338 pdpe = addr_gpa2hva(vm, pml4e[index[3]].pfn * vm->page_size); 339 TEST_ASSERT(pdpe[index[2]].present, 340 "Expected pdpe to be present for gva: 0x%08lx", vaddr); 341 TEST_ASSERT(pdpe[index[2]].page_size == 0, 342 "Expected pdpe to map a pde not a 1-GByte page."); 343 TEST_ASSERT((*(uint64_t*)(&pdpe[index[2]]) & rsvd_mask) == 0, 344 "Unexpected reserved bits set."); 345 346 pde = addr_gpa2hva(vm, pdpe[index[2]].pfn * vm->page_size); 347 TEST_ASSERT(pde[index[1]].present, 348 "Expected pde to be present for gva: 0x%08lx", vaddr); 349 TEST_ASSERT(pde[index[1]].page_size == 0, 350 "Expected pde to map a pte not a 2-MByte page."); 351 TEST_ASSERT((*(uint64_t*)(&pde[index[1]]) & rsvd_mask) == 0, 352 "Unexpected reserved bits set."); 353 354 pte = addr_gpa2hva(vm, pde[index[1]].pfn * vm->page_size); 355 TEST_ASSERT(pte[index[0]].present, 356 "Expected pte to be present for gva: 0x%08lx", vaddr); 357 358 return &pte[index[0]]; 359 } 360 361 uint64_t vm_get_page_table_entry(struct kvm_vm *vm, int vcpuid, uint64_t vaddr) 362 { 363 struct pageTableEntry *pte = _vm_get_page_table_entry(vm, vcpuid, vaddr); 364 365 return *(uint64_t *)pte; 366 } 367 368 void vm_set_page_table_entry(struct kvm_vm *vm, int vcpuid, uint64_t vaddr, 369 uint64_t pte) 370 { 371 struct pageTableEntry *new_pte = _vm_get_page_table_entry(vm, vcpuid, 372 vaddr); 373 374 *(uint64_t *)new_pte = pte; 375 } 376 377 void virt_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent) 378 { 379 struct pageUpperEntry *pml4e, *pml4e_start; 380 struct pageUpperEntry *pdpe, *pdpe_start; 381 struct pageUpperEntry *pde, *pde_start; 382 struct pageTableEntry *pte, *pte_start; 383 384 if (!vm->pgd_created) 385 return; 386 387 fprintf(stream, "%*s " 388 " no\n", indent, ""); 389 fprintf(stream, "%*s index hvaddr gpaddr " 390 "addr w exec dirty\n", 391 indent, ""); 392 pml4e_start = (struct pageUpperEntry *) addr_gpa2hva(vm, vm->pgd); 393 for (uint16_t n1 = 0; n1 <= 0x1ffu; n1++) { 394 pml4e = &pml4e_start[n1]; 395 if (!pml4e->present) 396 continue; 397 fprintf(stream, "%*spml4e 0x%-3zx %p 0x%-12lx 0x%-10lx %u " 398 " %u\n", 399 indent, "", 400 pml4e - pml4e_start, pml4e, 401 addr_hva2gpa(vm, pml4e), (uint64_t) pml4e->pfn, 402 pml4e->writable, pml4e->execute_disable); 403 404 pdpe_start = addr_gpa2hva(vm, pml4e->pfn * vm->page_size); 405 for (uint16_t n2 = 0; n2 <= 0x1ffu; n2++) { 406 pdpe = &pdpe_start[n2]; 407 if (!pdpe->present) 408 continue; 409 fprintf(stream, "%*spdpe 0x%-3zx %p 0x%-12lx 0x%-10lx " 410 "%u %u\n", 411 indent, "", 412 pdpe - pdpe_start, pdpe, 413 addr_hva2gpa(vm, pdpe), 414 (uint64_t) pdpe->pfn, pdpe->writable, 415 pdpe->execute_disable); 416 417 pde_start = addr_gpa2hva(vm, pdpe->pfn * vm->page_size); 418 for (uint16_t n3 = 0; n3 <= 0x1ffu; n3++) { 419 pde = &pde_start[n3]; 420 if (!pde->present) 421 continue; 422 fprintf(stream, "%*spde 0x%-3zx %p " 423 "0x%-12lx 0x%-10lx %u %u\n", 424 indent, "", pde - pde_start, pde, 425 addr_hva2gpa(vm, pde), 426 (uint64_t) pde->pfn, pde->writable, 427 pde->execute_disable); 428 429 pte_start = addr_gpa2hva(vm, pde->pfn * vm->page_size); 430 for (uint16_t n4 = 0; n4 <= 0x1ffu; n4++) { 431 pte = &pte_start[n4]; 432 if (!pte->present) 433 continue; 434 fprintf(stream, "%*spte 0x%-3zx %p " 435 "0x%-12lx 0x%-10lx %u %u " 436 " %u 0x%-10lx\n", 437 indent, "", 438 pte - pte_start, pte, 439 addr_hva2gpa(vm, pte), 440 (uint64_t) pte->pfn, 441 pte->writable, 442 pte->execute_disable, 443 pte->dirty, 444 ((uint64_t) n1 << 27) 445 | ((uint64_t) n2 << 18) 446 | ((uint64_t) n3 << 9) 447 | ((uint64_t) n4)); 448 } 449 } 450 } 451 } 452 } 453 454 /* 455 * Set Unusable Segment 456 * 457 * Input Args: None 458 * 459 * Output Args: 460 * segp - Pointer to segment register 461 * 462 * Return: None 463 * 464 * Sets the segment register pointed to by @segp to an unusable state. 465 */ 466 static void kvm_seg_set_unusable(struct kvm_segment *segp) 467 { 468 memset(segp, 0, sizeof(*segp)); 469 segp->unusable = true; 470 } 471 472 static void kvm_seg_fill_gdt_64bit(struct kvm_vm *vm, struct kvm_segment *segp) 473 { 474 void *gdt = addr_gva2hva(vm, vm->gdt); 475 struct desc64 *desc = gdt + (segp->selector >> 3) * 8; 476 477 desc->limit0 = segp->limit & 0xFFFF; 478 desc->base0 = segp->base & 0xFFFF; 479 desc->base1 = segp->base >> 16; 480 desc->type = segp->type; 481 desc->s = segp->s; 482 desc->dpl = segp->dpl; 483 desc->p = segp->present; 484 desc->limit1 = segp->limit >> 16; 485 desc->avl = segp->avl; 486 desc->l = segp->l; 487 desc->db = segp->db; 488 desc->g = segp->g; 489 desc->base2 = segp->base >> 24; 490 if (!segp->s) 491 desc->base3 = segp->base >> 32; 492 } 493 494 495 /* 496 * Set Long Mode Flat Kernel Code Segment 497 * 498 * Input Args: 499 * vm - VM whose GDT is being filled, or NULL to only write segp 500 * selector - selector value 501 * 502 * Output Args: 503 * segp - Pointer to KVM segment 504 * 505 * Return: None 506 * 507 * Sets up the KVM segment pointed to by @segp, to be a code segment 508 * with the selector value given by @selector. 509 */ 510 static void kvm_seg_set_kernel_code_64bit(struct kvm_vm *vm, uint16_t selector, 511 struct kvm_segment *segp) 512 { 513 memset(segp, 0, sizeof(*segp)); 514 segp->selector = selector; 515 segp->limit = 0xFFFFFFFFu; 516 segp->s = 0x1; /* kTypeCodeData */ 517 segp->type = 0x08 | 0x01 | 0x02; /* kFlagCode | kFlagCodeAccessed 518 * | kFlagCodeReadable 519 */ 520 segp->g = true; 521 segp->l = true; 522 segp->present = 1; 523 if (vm) 524 kvm_seg_fill_gdt_64bit(vm, segp); 525 } 526 527 /* 528 * Set Long Mode Flat Kernel Data Segment 529 * 530 * Input Args: 531 * vm - VM whose GDT is being filled, or NULL to only write segp 532 * selector - selector value 533 * 534 * Output Args: 535 * segp - Pointer to KVM segment 536 * 537 * Return: None 538 * 539 * Sets up the KVM segment pointed to by @segp, to be a data segment 540 * with the selector value given by @selector. 541 */ 542 static void kvm_seg_set_kernel_data_64bit(struct kvm_vm *vm, uint16_t selector, 543 struct kvm_segment *segp) 544 { 545 memset(segp, 0, sizeof(*segp)); 546 segp->selector = selector; 547 segp->limit = 0xFFFFFFFFu; 548 segp->s = 0x1; /* kTypeCodeData */ 549 segp->type = 0x00 | 0x01 | 0x02; /* kFlagData | kFlagDataAccessed 550 * | kFlagDataWritable 551 */ 552 segp->g = true; 553 segp->present = true; 554 if (vm) 555 kvm_seg_fill_gdt_64bit(vm, segp); 556 } 557 558 vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva) 559 { 560 uint16_t index[4]; 561 struct pageUpperEntry *pml4e, *pdpe, *pde; 562 struct pageTableEntry *pte; 563 564 TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, "Attempt to use " 565 "unknown or unsupported guest mode, mode: 0x%x", vm->mode); 566 567 index[0] = (gva >> 12) & 0x1ffu; 568 index[1] = (gva >> 21) & 0x1ffu; 569 index[2] = (gva >> 30) & 0x1ffu; 570 index[3] = (gva >> 39) & 0x1ffu; 571 572 if (!vm->pgd_created) 573 goto unmapped_gva; 574 pml4e = addr_gpa2hva(vm, vm->pgd); 575 if (!pml4e[index[3]].present) 576 goto unmapped_gva; 577 578 pdpe = addr_gpa2hva(vm, pml4e[index[3]].pfn * vm->page_size); 579 if (!pdpe[index[2]].present) 580 goto unmapped_gva; 581 582 pde = addr_gpa2hva(vm, pdpe[index[2]].pfn * vm->page_size); 583 if (!pde[index[1]].present) 584 goto unmapped_gva; 585 586 pte = addr_gpa2hva(vm, pde[index[1]].pfn * vm->page_size); 587 if (!pte[index[0]].present) 588 goto unmapped_gva; 589 590 return (pte[index[0]].pfn * vm->page_size) + (gva & 0xfffu); 591 592 unmapped_gva: 593 TEST_FAIL("No mapping for vm virtual address, gva: 0x%lx", gva); 594 exit(EXIT_FAILURE); 595 } 596 597 static void kvm_setup_gdt(struct kvm_vm *vm, struct kvm_dtable *dt) 598 { 599 if (!vm->gdt) 600 vm->gdt = vm_vaddr_alloc_page(vm); 601 602 dt->base = vm->gdt; 603 dt->limit = getpagesize(); 604 } 605 606 static void kvm_setup_tss_64bit(struct kvm_vm *vm, struct kvm_segment *segp, 607 int selector) 608 { 609 if (!vm->tss) 610 vm->tss = vm_vaddr_alloc_page(vm); 611 612 memset(segp, 0, sizeof(*segp)); 613 segp->base = vm->tss; 614 segp->limit = 0x67; 615 segp->selector = selector; 616 segp->type = 0xb; 617 segp->present = 1; 618 kvm_seg_fill_gdt_64bit(vm, segp); 619 } 620 621 static void vcpu_setup(struct kvm_vm *vm, int vcpuid) 622 { 623 struct kvm_sregs sregs; 624 625 /* Set mode specific system register values. */ 626 vcpu_sregs_get(vm, vcpuid, &sregs); 627 628 sregs.idt.limit = 0; 629 630 kvm_setup_gdt(vm, &sregs.gdt); 631 632 switch (vm->mode) { 633 case VM_MODE_PXXV48_4K: 634 sregs.cr0 = X86_CR0_PE | X86_CR0_NE | X86_CR0_PG; 635 sregs.cr4 |= X86_CR4_PAE | X86_CR4_OSFXSR; 636 sregs.efer |= (EFER_LME | EFER_LMA | EFER_NX); 637 638 kvm_seg_set_unusable(&sregs.ldt); 639 kvm_seg_set_kernel_code_64bit(vm, DEFAULT_CODE_SELECTOR, &sregs.cs); 640 kvm_seg_set_kernel_data_64bit(vm, DEFAULT_DATA_SELECTOR, &sregs.ds); 641 kvm_seg_set_kernel_data_64bit(vm, DEFAULT_DATA_SELECTOR, &sregs.es); 642 kvm_setup_tss_64bit(vm, &sregs.tr, 0x18); 643 break; 644 645 default: 646 TEST_FAIL("Unknown guest mode, mode: 0x%x", vm->mode); 647 } 648 649 sregs.cr3 = vm->pgd; 650 vcpu_sregs_set(vm, vcpuid, &sregs); 651 } 652 653 void vm_vcpu_add_default(struct kvm_vm *vm, uint32_t vcpuid, void *guest_code) 654 { 655 struct kvm_mp_state mp_state; 656 struct kvm_regs regs; 657 vm_vaddr_t stack_vaddr; 658 stack_vaddr = vm_vaddr_alloc(vm, DEFAULT_STACK_PGS * getpagesize(), 659 DEFAULT_GUEST_STACK_VADDR_MIN); 660 661 /* Create VCPU */ 662 vm_vcpu_add(vm, vcpuid); 663 vcpu_set_cpuid(vm, vcpuid, kvm_get_supported_cpuid()); 664 vcpu_setup(vm, vcpuid); 665 666 /* Setup guest general purpose registers */ 667 vcpu_regs_get(vm, vcpuid, ®s); 668 regs.rflags = regs.rflags | 0x2; 669 regs.rsp = stack_vaddr + (DEFAULT_STACK_PGS * getpagesize()); 670 regs.rip = (unsigned long) guest_code; 671 vcpu_regs_set(vm, vcpuid, ®s); 672 673 /* Setup the MP state */ 674 mp_state.mp_state = 0; 675 vcpu_set_mp_state(vm, vcpuid, &mp_state); 676 } 677 678 /* 679 * Allocate an instance of struct kvm_cpuid2 680 * 681 * Input Args: None 682 * 683 * Output Args: None 684 * 685 * Return: A pointer to the allocated struct. The caller is responsible 686 * for freeing this struct. 687 * 688 * Since kvm_cpuid2 uses a 0-length array to allow a the size of the 689 * array to be decided at allocation time, allocation is slightly 690 * complicated. This function uses a reasonable default length for 691 * the array and performs the appropriate allocation. 692 */ 693 static struct kvm_cpuid2 *allocate_kvm_cpuid2(void) 694 { 695 struct kvm_cpuid2 *cpuid; 696 int nent = 100; 697 size_t size; 698 699 size = sizeof(*cpuid); 700 size += nent * sizeof(struct kvm_cpuid_entry2); 701 cpuid = malloc(size); 702 if (!cpuid) { 703 perror("malloc"); 704 abort(); 705 } 706 707 cpuid->nent = nent; 708 709 return cpuid; 710 } 711 712 /* 713 * KVM Supported CPUID Get 714 * 715 * Input Args: None 716 * 717 * Output Args: 718 * 719 * Return: The supported KVM CPUID 720 * 721 * Get the guest CPUID supported by KVM. 722 */ 723 struct kvm_cpuid2 *kvm_get_supported_cpuid(void) 724 { 725 static struct kvm_cpuid2 *cpuid; 726 int ret; 727 int kvm_fd; 728 729 if (cpuid) 730 return cpuid; 731 732 cpuid = allocate_kvm_cpuid2(); 733 kvm_fd = open_kvm_dev_path_or_exit(); 734 735 ret = ioctl(kvm_fd, KVM_GET_SUPPORTED_CPUID, cpuid); 736 TEST_ASSERT(ret == 0, "KVM_GET_SUPPORTED_CPUID failed %d %d\n", 737 ret, errno); 738 739 close(kvm_fd); 740 return cpuid; 741 } 742 743 /* 744 * KVM Get MSR 745 * 746 * Input Args: 747 * msr_index - Index of MSR 748 * 749 * Output Args: None 750 * 751 * Return: On success, value of the MSR. On failure a TEST_ASSERT is produced. 752 * 753 * Get value of MSR for VCPU. 754 */ 755 uint64_t kvm_get_feature_msr(uint64_t msr_index) 756 { 757 struct { 758 struct kvm_msrs header; 759 struct kvm_msr_entry entry; 760 } buffer = {}; 761 int r, kvm_fd; 762 763 buffer.header.nmsrs = 1; 764 buffer.entry.index = msr_index; 765 kvm_fd = open_kvm_dev_path_or_exit(); 766 767 r = ioctl(kvm_fd, KVM_GET_MSRS, &buffer.header); 768 TEST_ASSERT(r == 1, "KVM_GET_MSRS IOCTL failed,\n" 769 " rc: %i errno: %i", r, errno); 770 771 close(kvm_fd); 772 return buffer.entry.data; 773 } 774 775 /* 776 * VM VCPU CPUID Set 777 * 778 * Input Args: 779 * vm - Virtual Machine 780 * vcpuid - VCPU id 781 * 782 * Output Args: None 783 * 784 * Return: KVM CPUID (KVM_GET_CPUID2) 785 * 786 * Set the VCPU's CPUID. 787 */ 788 struct kvm_cpuid2 *vcpu_get_cpuid(struct kvm_vm *vm, uint32_t vcpuid) 789 { 790 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 791 struct kvm_cpuid2 *cpuid; 792 int max_ent; 793 int rc = -1; 794 795 TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid); 796 797 cpuid = allocate_kvm_cpuid2(); 798 max_ent = cpuid->nent; 799 800 for (cpuid->nent = 1; cpuid->nent <= max_ent; cpuid->nent++) { 801 rc = ioctl(vcpu->fd, KVM_GET_CPUID2, cpuid); 802 if (!rc) 803 break; 804 805 TEST_ASSERT(rc == -1 && errno == E2BIG, 806 "KVM_GET_CPUID2 should either succeed or give E2BIG: %d %d", 807 rc, errno); 808 } 809 810 TEST_ASSERT(rc == 0, "KVM_GET_CPUID2 failed, rc: %i errno: %i", 811 rc, errno); 812 813 return cpuid; 814 } 815 816 817 818 /* 819 * Locate a cpuid entry. 820 * 821 * Input Args: 822 * function: The function of the cpuid entry to find. 823 * index: The index of the cpuid entry. 824 * 825 * Output Args: None 826 * 827 * Return: A pointer to the cpuid entry. Never returns NULL. 828 */ 829 struct kvm_cpuid_entry2 * 830 kvm_get_supported_cpuid_index(uint32_t function, uint32_t index) 831 { 832 struct kvm_cpuid2 *cpuid; 833 struct kvm_cpuid_entry2 *entry = NULL; 834 int i; 835 836 cpuid = kvm_get_supported_cpuid(); 837 for (i = 0; i < cpuid->nent; i++) { 838 if (cpuid->entries[i].function == function && 839 cpuid->entries[i].index == index) { 840 entry = &cpuid->entries[i]; 841 break; 842 } 843 } 844 845 TEST_ASSERT(entry, "Guest CPUID entry not found: (EAX=%x, ECX=%x).", 846 function, index); 847 return entry; 848 } 849 850 /* 851 * VM VCPU CPUID Set 852 * 853 * Input Args: 854 * vm - Virtual Machine 855 * vcpuid - VCPU id 856 * cpuid - The CPUID values to set. 857 * 858 * Output Args: None 859 * 860 * Return: void 861 * 862 * Set the VCPU's CPUID. 863 */ 864 void vcpu_set_cpuid(struct kvm_vm *vm, 865 uint32_t vcpuid, struct kvm_cpuid2 *cpuid) 866 { 867 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 868 int rc; 869 870 TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid); 871 872 rc = ioctl(vcpu->fd, KVM_SET_CPUID2, cpuid); 873 TEST_ASSERT(rc == 0, "KVM_SET_CPUID2 failed, rc: %i errno: %i", 874 rc, errno); 875 876 } 877 878 /* 879 * VCPU Get MSR 880 * 881 * Input Args: 882 * vm - Virtual Machine 883 * vcpuid - VCPU ID 884 * msr_index - Index of MSR 885 * 886 * Output Args: None 887 * 888 * Return: On success, value of the MSR. On failure a TEST_ASSERT is produced. 889 * 890 * Get value of MSR for VCPU. 891 */ 892 uint64_t vcpu_get_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index) 893 { 894 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 895 struct { 896 struct kvm_msrs header; 897 struct kvm_msr_entry entry; 898 } buffer = {}; 899 int r; 900 901 TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid); 902 buffer.header.nmsrs = 1; 903 buffer.entry.index = msr_index; 904 r = ioctl(vcpu->fd, KVM_GET_MSRS, &buffer.header); 905 TEST_ASSERT(r == 1, "KVM_GET_MSRS IOCTL failed,\n" 906 " rc: %i errno: %i", r, errno); 907 908 return buffer.entry.data; 909 } 910 911 /* 912 * _VCPU Set MSR 913 * 914 * Input Args: 915 * vm - Virtual Machine 916 * vcpuid - VCPU ID 917 * msr_index - Index of MSR 918 * msr_value - New value of MSR 919 * 920 * Output Args: None 921 * 922 * Return: The result of KVM_SET_MSRS. 923 * 924 * Sets the value of an MSR for the given VCPU. 925 */ 926 int _vcpu_set_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index, 927 uint64_t msr_value) 928 { 929 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 930 struct { 931 struct kvm_msrs header; 932 struct kvm_msr_entry entry; 933 } buffer = {}; 934 int r; 935 936 TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid); 937 memset(&buffer, 0, sizeof(buffer)); 938 buffer.header.nmsrs = 1; 939 buffer.entry.index = msr_index; 940 buffer.entry.data = msr_value; 941 r = ioctl(vcpu->fd, KVM_SET_MSRS, &buffer.header); 942 return r; 943 } 944 945 /* 946 * VCPU Set MSR 947 * 948 * Input Args: 949 * vm - Virtual Machine 950 * vcpuid - VCPU ID 951 * msr_index - Index of MSR 952 * msr_value - New value of MSR 953 * 954 * Output Args: None 955 * 956 * Return: On success, nothing. On failure a TEST_ASSERT is produced. 957 * 958 * Set value of MSR for VCPU. 959 */ 960 void vcpu_set_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index, 961 uint64_t msr_value) 962 { 963 int r; 964 965 r = _vcpu_set_msr(vm, vcpuid, msr_index, msr_value); 966 TEST_ASSERT(r == 1, "KVM_SET_MSRS IOCTL failed,\n" 967 " rc: %i errno: %i", r, errno); 968 } 969 970 void vcpu_args_set(struct kvm_vm *vm, uint32_t vcpuid, unsigned int num, ...) 971 { 972 va_list ap; 973 struct kvm_regs regs; 974 975 TEST_ASSERT(num >= 1 && num <= 6, "Unsupported number of args,\n" 976 " num: %u\n", 977 num); 978 979 va_start(ap, num); 980 vcpu_regs_get(vm, vcpuid, ®s); 981 982 if (num >= 1) 983 regs.rdi = va_arg(ap, uint64_t); 984 985 if (num >= 2) 986 regs.rsi = va_arg(ap, uint64_t); 987 988 if (num >= 3) 989 regs.rdx = va_arg(ap, uint64_t); 990 991 if (num >= 4) 992 regs.rcx = va_arg(ap, uint64_t); 993 994 if (num >= 5) 995 regs.r8 = va_arg(ap, uint64_t); 996 997 if (num >= 6) 998 regs.r9 = va_arg(ap, uint64_t); 999 1000 vcpu_regs_set(vm, vcpuid, ®s); 1001 va_end(ap); 1002 } 1003 1004 void vcpu_dump(FILE *stream, struct kvm_vm *vm, uint32_t vcpuid, uint8_t indent) 1005 { 1006 struct kvm_regs regs; 1007 struct kvm_sregs sregs; 1008 1009 fprintf(stream, "%*scpuid: %u\n", indent, "", vcpuid); 1010 1011 fprintf(stream, "%*sregs:\n", indent + 2, ""); 1012 vcpu_regs_get(vm, vcpuid, ®s); 1013 regs_dump(stream, ®s, indent + 4); 1014 1015 fprintf(stream, "%*ssregs:\n", indent + 2, ""); 1016 vcpu_sregs_get(vm, vcpuid, &sregs); 1017 sregs_dump(stream, &sregs, indent + 4); 1018 } 1019 1020 struct kvm_x86_state { 1021 struct kvm_vcpu_events events; 1022 struct kvm_mp_state mp_state; 1023 struct kvm_regs regs; 1024 struct kvm_xsave xsave; 1025 struct kvm_xcrs xcrs; 1026 struct kvm_sregs sregs; 1027 struct kvm_debugregs debugregs; 1028 union { 1029 struct kvm_nested_state nested; 1030 char nested_[16384]; 1031 }; 1032 struct kvm_msrs msrs; 1033 }; 1034 1035 static int kvm_get_num_msrs_fd(int kvm_fd) 1036 { 1037 struct kvm_msr_list nmsrs; 1038 int r; 1039 1040 nmsrs.nmsrs = 0; 1041 r = ioctl(kvm_fd, KVM_GET_MSR_INDEX_LIST, &nmsrs); 1042 TEST_ASSERT(r == -1 && errno == E2BIG, "Unexpected result from KVM_GET_MSR_INDEX_LIST probe, r: %i", 1043 r); 1044 1045 return nmsrs.nmsrs; 1046 } 1047 1048 static int kvm_get_num_msrs(struct kvm_vm *vm) 1049 { 1050 return kvm_get_num_msrs_fd(vm->kvm_fd); 1051 } 1052 1053 struct kvm_msr_list *kvm_get_msr_index_list(void) 1054 { 1055 struct kvm_msr_list *list; 1056 int nmsrs, r, kvm_fd; 1057 1058 kvm_fd = open_kvm_dev_path_or_exit(); 1059 1060 nmsrs = kvm_get_num_msrs_fd(kvm_fd); 1061 list = malloc(sizeof(*list) + nmsrs * sizeof(list->indices[0])); 1062 list->nmsrs = nmsrs; 1063 r = ioctl(kvm_fd, KVM_GET_MSR_INDEX_LIST, list); 1064 close(kvm_fd); 1065 1066 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MSR_INDEX_LIST, r: %i", 1067 r); 1068 1069 return list; 1070 } 1071 1072 struct kvm_x86_state *vcpu_save_state(struct kvm_vm *vm, uint32_t vcpuid) 1073 { 1074 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 1075 struct kvm_msr_list *list; 1076 struct kvm_x86_state *state; 1077 int nmsrs, r, i; 1078 static int nested_size = -1; 1079 1080 if (nested_size == -1) { 1081 nested_size = kvm_check_cap(KVM_CAP_NESTED_STATE); 1082 TEST_ASSERT(nested_size <= sizeof(state->nested_), 1083 "Nested state size too big, %i > %zi", 1084 nested_size, sizeof(state->nested_)); 1085 } 1086 1087 /* 1088 * When KVM exits to userspace with KVM_EXIT_IO, KVM guarantees 1089 * guest state is consistent only after userspace re-enters the 1090 * kernel with KVM_RUN. Complete IO prior to migrating state 1091 * to a new VM. 1092 */ 1093 vcpu_run_complete_io(vm, vcpuid); 1094 1095 nmsrs = kvm_get_num_msrs(vm); 1096 list = malloc(sizeof(*list) + nmsrs * sizeof(list->indices[0])); 1097 list->nmsrs = nmsrs; 1098 r = ioctl(vm->kvm_fd, KVM_GET_MSR_INDEX_LIST, list); 1099 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MSR_INDEX_LIST, r: %i", 1100 r); 1101 1102 state = malloc(sizeof(*state) + nmsrs * sizeof(state->msrs.entries[0])); 1103 r = ioctl(vcpu->fd, KVM_GET_VCPU_EVENTS, &state->events); 1104 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_VCPU_EVENTS, r: %i", 1105 r); 1106 1107 r = ioctl(vcpu->fd, KVM_GET_MP_STATE, &state->mp_state); 1108 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MP_STATE, r: %i", 1109 r); 1110 1111 r = ioctl(vcpu->fd, KVM_GET_REGS, &state->regs); 1112 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_REGS, r: %i", 1113 r); 1114 1115 r = ioctl(vcpu->fd, KVM_GET_XSAVE, &state->xsave); 1116 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_XSAVE, r: %i", 1117 r); 1118 1119 if (kvm_check_cap(KVM_CAP_XCRS)) { 1120 r = ioctl(vcpu->fd, KVM_GET_XCRS, &state->xcrs); 1121 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_XCRS, r: %i", 1122 r); 1123 } 1124 1125 r = ioctl(vcpu->fd, KVM_GET_SREGS, &state->sregs); 1126 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_SREGS, r: %i", 1127 r); 1128 1129 if (nested_size) { 1130 state->nested.size = sizeof(state->nested_); 1131 r = ioctl(vcpu->fd, KVM_GET_NESTED_STATE, &state->nested); 1132 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_NESTED_STATE, r: %i", 1133 r); 1134 TEST_ASSERT(state->nested.size <= nested_size, 1135 "Nested state size too big, %i (KVM_CHECK_CAP gave %i)", 1136 state->nested.size, nested_size); 1137 } else 1138 state->nested.size = 0; 1139 1140 state->msrs.nmsrs = nmsrs; 1141 for (i = 0; i < nmsrs; i++) 1142 state->msrs.entries[i].index = list->indices[i]; 1143 r = ioctl(vcpu->fd, KVM_GET_MSRS, &state->msrs); 1144 TEST_ASSERT(r == nmsrs, "Unexpected result from KVM_GET_MSRS, r: %i (failed MSR was 0x%x)", 1145 r, r == nmsrs ? -1 : list->indices[r]); 1146 1147 r = ioctl(vcpu->fd, KVM_GET_DEBUGREGS, &state->debugregs); 1148 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_DEBUGREGS, r: %i", 1149 r); 1150 1151 free(list); 1152 return state; 1153 } 1154 1155 void vcpu_load_state(struct kvm_vm *vm, uint32_t vcpuid, struct kvm_x86_state *state) 1156 { 1157 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 1158 int r; 1159 1160 r = ioctl(vcpu->fd, KVM_SET_XSAVE, &state->xsave); 1161 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_XSAVE, r: %i", 1162 r); 1163 1164 if (kvm_check_cap(KVM_CAP_XCRS)) { 1165 r = ioctl(vcpu->fd, KVM_SET_XCRS, &state->xcrs); 1166 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_XCRS, r: %i", 1167 r); 1168 } 1169 1170 r = ioctl(vcpu->fd, KVM_SET_SREGS, &state->sregs); 1171 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_SREGS, r: %i", 1172 r); 1173 1174 r = ioctl(vcpu->fd, KVM_SET_MSRS, &state->msrs); 1175 TEST_ASSERT(r == state->msrs.nmsrs, "Unexpected result from KVM_SET_MSRS, r: %i (failed at %x)", 1176 r, r == state->msrs.nmsrs ? -1 : state->msrs.entries[r].index); 1177 1178 r = ioctl(vcpu->fd, KVM_SET_VCPU_EVENTS, &state->events); 1179 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_VCPU_EVENTS, r: %i", 1180 r); 1181 1182 r = ioctl(vcpu->fd, KVM_SET_MP_STATE, &state->mp_state); 1183 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_MP_STATE, r: %i", 1184 r); 1185 1186 r = ioctl(vcpu->fd, KVM_SET_DEBUGREGS, &state->debugregs); 1187 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_DEBUGREGS, r: %i", 1188 r); 1189 1190 r = ioctl(vcpu->fd, KVM_SET_REGS, &state->regs); 1191 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_REGS, r: %i", 1192 r); 1193 1194 if (state->nested.size) { 1195 r = ioctl(vcpu->fd, KVM_SET_NESTED_STATE, &state->nested); 1196 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_NESTED_STATE, r: %i", 1197 r); 1198 } 1199 } 1200 1201 bool is_intel_cpu(void) 1202 { 1203 int eax, ebx, ecx, edx; 1204 const uint32_t *chunk; 1205 const int leaf = 0; 1206 1207 __asm__ __volatile__( 1208 "cpuid" 1209 : /* output */ "=a"(eax), "=b"(ebx), 1210 "=c"(ecx), "=d"(edx) 1211 : /* input */ "0"(leaf), "2"(0)); 1212 1213 chunk = (const uint32_t *)("GenuineIntel"); 1214 return (ebx == chunk[0] && edx == chunk[1] && ecx == chunk[2]); 1215 } 1216 1217 uint32_t kvm_get_cpuid_max_basic(void) 1218 { 1219 return kvm_get_supported_cpuid_entry(0)->eax; 1220 } 1221 1222 uint32_t kvm_get_cpuid_max_extended(void) 1223 { 1224 return kvm_get_supported_cpuid_entry(0x80000000)->eax; 1225 } 1226 1227 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits) 1228 { 1229 struct kvm_cpuid_entry2 *entry; 1230 bool pae; 1231 1232 /* SDM 4.1.4 */ 1233 if (kvm_get_cpuid_max_extended() < 0x80000008) { 1234 pae = kvm_get_supported_cpuid_entry(1)->edx & (1 << 6); 1235 *pa_bits = pae ? 36 : 32; 1236 *va_bits = 32; 1237 } else { 1238 entry = kvm_get_supported_cpuid_entry(0x80000008); 1239 *pa_bits = entry->eax & 0xff; 1240 *va_bits = (entry->eax >> 8) & 0xff; 1241 } 1242 } 1243 1244 struct idt_entry { 1245 uint16_t offset0; 1246 uint16_t selector; 1247 uint16_t ist : 3; 1248 uint16_t : 5; 1249 uint16_t type : 4; 1250 uint16_t : 1; 1251 uint16_t dpl : 2; 1252 uint16_t p : 1; 1253 uint16_t offset1; 1254 uint32_t offset2; uint32_t reserved; 1255 }; 1256 1257 static void set_idt_entry(struct kvm_vm *vm, int vector, unsigned long addr, 1258 int dpl, unsigned short selector) 1259 { 1260 struct idt_entry *base = 1261 (struct idt_entry *)addr_gva2hva(vm, vm->idt); 1262 struct idt_entry *e = &base[vector]; 1263 1264 memset(e, 0, sizeof(*e)); 1265 e->offset0 = addr; 1266 e->selector = selector; 1267 e->ist = 0; 1268 e->type = 14; 1269 e->dpl = dpl; 1270 e->p = 1; 1271 e->offset1 = addr >> 16; 1272 e->offset2 = addr >> 32; 1273 } 1274 1275 void kvm_exit_unexpected_vector(uint32_t value) 1276 { 1277 ucall(UCALL_UNHANDLED, 1, value); 1278 } 1279 1280 void route_exception(struct ex_regs *regs) 1281 { 1282 typedef void(*handler)(struct ex_regs *); 1283 handler *handlers = (handler *)exception_handlers; 1284 1285 if (handlers && handlers[regs->vector]) { 1286 handlers[regs->vector](regs); 1287 return; 1288 } 1289 1290 kvm_exit_unexpected_vector(regs->vector); 1291 } 1292 1293 void vm_init_descriptor_tables(struct kvm_vm *vm) 1294 { 1295 extern void *idt_handlers; 1296 int i; 1297 1298 vm->idt = vm_vaddr_alloc_page(vm); 1299 vm->handlers = vm_vaddr_alloc_page(vm); 1300 /* Handlers have the same address in both address spaces.*/ 1301 for (i = 0; i < NUM_INTERRUPTS; i++) 1302 set_idt_entry(vm, i, (unsigned long)(&idt_handlers)[i], 0, 1303 DEFAULT_CODE_SELECTOR); 1304 } 1305 1306 void vcpu_init_descriptor_tables(struct kvm_vm *vm, uint32_t vcpuid) 1307 { 1308 struct kvm_sregs sregs; 1309 1310 vcpu_sregs_get(vm, vcpuid, &sregs); 1311 sregs.idt.base = vm->idt; 1312 sregs.idt.limit = NUM_INTERRUPTS * sizeof(struct idt_entry) - 1; 1313 sregs.gdt.base = vm->gdt; 1314 sregs.gdt.limit = getpagesize() - 1; 1315 kvm_seg_set_kernel_data_64bit(NULL, DEFAULT_DATA_SELECTOR, &sregs.gs); 1316 vcpu_sregs_set(vm, vcpuid, &sregs); 1317 *(vm_vaddr_t *)addr_gva2hva(vm, (vm_vaddr_t)(&exception_handlers)) = vm->handlers; 1318 } 1319 1320 void vm_install_exception_handler(struct kvm_vm *vm, int vector, 1321 void (*handler)(struct ex_regs *)) 1322 { 1323 vm_vaddr_t *handlers = (vm_vaddr_t *)addr_gva2hva(vm, vm->handlers); 1324 1325 handlers[vector] = (vm_vaddr_t)handler; 1326 } 1327 1328 void assert_on_unhandled_exception(struct kvm_vm *vm, uint32_t vcpuid) 1329 { 1330 struct ucall uc; 1331 1332 if (get_ucall(vm, vcpuid, &uc) == UCALL_UNHANDLED) { 1333 uint64_t vector = uc.args[0]; 1334 1335 TEST_FAIL("Unexpected vectored event in guest (vector:0x%lx)", 1336 vector); 1337 } 1338 } 1339 1340 bool set_cpuid(struct kvm_cpuid2 *cpuid, 1341 struct kvm_cpuid_entry2 *ent) 1342 { 1343 int i; 1344 1345 for (i = 0; i < cpuid->nent; i++) { 1346 struct kvm_cpuid_entry2 *cur = &cpuid->entries[i]; 1347 1348 if (cur->function != ent->function || cur->index != ent->index) 1349 continue; 1350 1351 memcpy(cur, ent, sizeof(struct kvm_cpuid_entry2)); 1352 return true; 1353 } 1354 1355 return false; 1356 } 1357 1358 uint64_t kvm_hypercall(uint64_t nr, uint64_t a0, uint64_t a1, uint64_t a2, 1359 uint64_t a3) 1360 { 1361 uint64_t r; 1362 1363 asm volatile("vmcall" 1364 : "=a"(r) 1365 : "b"(a0), "c"(a1), "d"(a2), "S"(a3)); 1366 return r; 1367 } 1368 1369 struct kvm_cpuid2 *kvm_get_supported_hv_cpuid(void) 1370 { 1371 static struct kvm_cpuid2 *cpuid; 1372 int ret; 1373 int kvm_fd; 1374 1375 if (cpuid) 1376 return cpuid; 1377 1378 cpuid = allocate_kvm_cpuid2(); 1379 kvm_fd = open_kvm_dev_path_or_exit(); 1380 1381 ret = ioctl(kvm_fd, KVM_GET_SUPPORTED_HV_CPUID, cpuid); 1382 TEST_ASSERT(ret == 0, "KVM_GET_SUPPORTED_HV_CPUID failed %d %d\n", 1383 ret, errno); 1384 1385 close(kvm_fd); 1386 return cpuid; 1387 } 1388 1389 void vcpu_set_hv_cpuid(struct kvm_vm *vm, uint32_t vcpuid) 1390 { 1391 static struct kvm_cpuid2 *cpuid_full; 1392 struct kvm_cpuid2 *cpuid_sys, *cpuid_hv; 1393 int i, nent = 0; 1394 1395 if (!cpuid_full) { 1396 cpuid_sys = kvm_get_supported_cpuid(); 1397 cpuid_hv = kvm_get_supported_hv_cpuid(); 1398 1399 cpuid_full = malloc(sizeof(*cpuid_full) + 1400 (cpuid_sys->nent + cpuid_hv->nent) * 1401 sizeof(struct kvm_cpuid_entry2)); 1402 if (!cpuid_full) { 1403 perror("malloc"); 1404 abort(); 1405 } 1406 1407 /* Need to skip KVM CPUID leaves 0x400000xx */ 1408 for (i = 0; i < cpuid_sys->nent; i++) { 1409 if (cpuid_sys->entries[i].function >= 0x40000000 && 1410 cpuid_sys->entries[i].function < 0x40000100) 1411 continue; 1412 cpuid_full->entries[nent] = cpuid_sys->entries[i]; 1413 nent++; 1414 } 1415 1416 memcpy(&cpuid_full->entries[nent], cpuid_hv->entries, 1417 cpuid_hv->nent * sizeof(struct kvm_cpuid_entry2)); 1418 cpuid_full->nent = nent + cpuid_hv->nent; 1419 } 1420 1421 vcpu_set_cpuid(vm, vcpuid, cpuid_full); 1422 } 1423 1424 struct kvm_cpuid2 *vcpu_get_supported_hv_cpuid(struct kvm_vm *vm, uint32_t vcpuid) 1425 { 1426 static struct kvm_cpuid2 *cpuid; 1427 1428 cpuid = allocate_kvm_cpuid2(); 1429 1430 vcpu_ioctl(vm, vcpuid, KVM_GET_SUPPORTED_HV_CPUID, cpuid); 1431 1432 return cpuid; 1433 } 1434 1435 #define X86EMUL_CPUID_VENDOR_AuthenticAMD_ebx 0x68747541 1436 #define X86EMUL_CPUID_VENDOR_AuthenticAMD_ecx 0x444d4163 1437 #define X86EMUL_CPUID_VENDOR_AuthenticAMD_edx 0x69746e65 1438 1439 static inline unsigned x86_family(unsigned int eax) 1440 { 1441 unsigned int x86; 1442 1443 x86 = (eax >> 8) & 0xf; 1444 1445 if (x86 == 0xf) 1446 x86 += (eax >> 20) & 0xff; 1447 1448 return x86; 1449 } 1450 1451 unsigned long vm_compute_max_gfn(struct kvm_vm *vm) 1452 { 1453 const unsigned long num_ht_pages = 12 << (30 - vm->page_shift); /* 12 GiB */ 1454 unsigned long ht_gfn, max_gfn, max_pfn; 1455 uint32_t eax, ebx, ecx, edx, max_ext_leaf; 1456 1457 max_gfn = (1ULL << (vm->pa_bits - vm->page_shift)) - 1; 1458 1459 /* Avoid reserved HyperTransport region on AMD processors. */ 1460 eax = ecx = 0; 1461 cpuid(&eax, &ebx, &ecx, &edx); 1462 if (ebx != X86EMUL_CPUID_VENDOR_AuthenticAMD_ebx || 1463 ecx != X86EMUL_CPUID_VENDOR_AuthenticAMD_ecx || 1464 edx != X86EMUL_CPUID_VENDOR_AuthenticAMD_edx) 1465 return max_gfn; 1466 1467 /* On parts with <40 physical address bits, the area is fully hidden */ 1468 if (vm->pa_bits < 40) 1469 return max_gfn; 1470 1471 /* Before family 17h, the HyperTransport area is just below 1T. */ 1472 ht_gfn = (1 << 28) - num_ht_pages; 1473 eax = 1; 1474 cpuid(&eax, &ebx, &ecx, &edx); 1475 if (x86_family(eax) < 0x17) 1476 goto done; 1477 1478 /* 1479 * Otherwise it's at the top of the physical address space, possibly 1480 * reduced due to SME by bits 11:6 of CPUID[0x8000001f].EBX. Use 1481 * the old conservative value if MAXPHYADDR is not enumerated. 1482 */ 1483 eax = 0x80000000; 1484 cpuid(&eax, &ebx, &ecx, &edx); 1485 max_ext_leaf = eax; 1486 if (max_ext_leaf < 0x80000008) 1487 goto done; 1488 1489 eax = 0x80000008; 1490 cpuid(&eax, &ebx, &ecx, &edx); 1491 max_pfn = (1ULL << ((eax & 0xff) - vm->page_shift)) - 1; 1492 if (max_ext_leaf >= 0x8000001f) { 1493 eax = 0x8000001f; 1494 cpuid(&eax, &ebx, &ecx, &edx); 1495 max_pfn >>= (ebx >> 6) & 0x3f; 1496 } 1497 1498 ht_gfn = max_pfn - num_ht_pages; 1499 done: 1500 return min(max_gfn, ht_gfn - 1); 1501 } 1502