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 void regs_dump(FILE *stream, struct kvm_regs *regs, 23 uint8_t indent) 24 { 25 fprintf(stream, "%*srax: 0x%.16llx rbx: 0x%.16llx " 26 "rcx: 0x%.16llx rdx: 0x%.16llx\n", 27 indent, "", 28 regs->rax, regs->rbx, regs->rcx, regs->rdx); 29 fprintf(stream, "%*srsi: 0x%.16llx rdi: 0x%.16llx " 30 "rsp: 0x%.16llx rbp: 0x%.16llx\n", 31 indent, "", 32 regs->rsi, regs->rdi, regs->rsp, regs->rbp); 33 fprintf(stream, "%*sr8: 0x%.16llx r9: 0x%.16llx " 34 "r10: 0x%.16llx r11: 0x%.16llx\n", 35 indent, "", 36 regs->r8, regs->r9, regs->r10, regs->r11); 37 fprintf(stream, "%*sr12: 0x%.16llx r13: 0x%.16llx " 38 "r14: 0x%.16llx r15: 0x%.16llx\n", 39 indent, "", 40 regs->r12, regs->r13, regs->r14, regs->r15); 41 fprintf(stream, "%*srip: 0x%.16llx rfl: 0x%.16llx\n", 42 indent, "", 43 regs->rip, regs->rflags); 44 } 45 46 /* 47 * Segment Dump 48 * 49 * Input Args: 50 * stream - Output FILE stream 51 * segment - KVM segment 52 * indent - Left margin indent amount 53 * 54 * Output Args: None 55 * 56 * Return: None 57 * 58 * Dumps the state of the KVM segment given by @segment, to the FILE stream 59 * given by @stream. 60 */ 61 static void segment_dump(FILE *stream, struct kvm_segment *segment, 62 uint8_t indent) 63 { 64 fprintf(stream, "%*sbase: 0x%.16llx limit: 0x%.8x " 65 "selector: 0x%.4x type: 0x%.2x\n", 66 indent, "", segment->base, segment->limit, 67 segment->selector, segment->type); 68 fprintf(stream, "%*spresent: 0x%.2x dpl: 0x%.2x " 69 "db: 0x%.2x s: 0x%.2x l: 0x%.2x\n", 70 indent, "", segment->present, segment->dpl, 71 segment->db, segment->s, segment->l); 72 fprintf(stream, "%*sg: 0x%.2x avl: 0x%.2x " 73 "unusable: 0x%.2x padding: 0x%.2x\n", 74 indent, "", segment->g, segment->avl, 75 segment->unusable, segment->padding); 76 } 77 78 /* 79 * dtable Dump 80 * 81 * Input Args: 82 * stream - Output FILE stream 83 * dtable - KVM dtable 84 * indent - Left margin indent amount 85 * 86 * Output Args: None 87 * 88 * Return: None 89 * 90 * Dumps the state of the KVM dtable given by @dtable, to the FILE stream 91 * given by @stream. 92 */ 93 static void dtable_dump(FILE *stream, struct kvm_dtable *dtable, 94 uint8_t indent) 95 { 96 fprintf(stream, "%*sbase: 0x%.16llx limit: 0x%.4x " 97 "padding: 0x%.4x 0x%.4x 0x%.4x\n", 98 indent, "", dtable->base, dtable->limit, 99 dtable->padding[0], dtable->padding[1], dtable->padding[2]); 100 } 101 102 void sregs_dump(FILE *stream, struct kvm_sregs *sregs, 103 uint8_t indent) 104 { 105 unsigned int i; 106 107 fprintf(stream, "%*scs:\n", indent, ""); 108 segment_dump(stream, &sregs->cs, indent + 2); 109 fprintf(stream, "%*sds:\n", indent, ""); 110 segment_dump(stream, &sregs->ds, indent + 2); 111 fprintf(stream, "%*ses:\n", indent, ""); 112 segment_dump(stream, &sregs->es, indent + 2); 113 fprintf(stream, "%*sfs:\n", indent, ""); 114 segment_dump(stream, &sregs->fs, indent + 2); 115 fprintf(stream, "%*sgs:\n", indent, ""); 116 segment_dump(stream, &sregs->gs, indent + 2); 117 fprintf(stream, "%*sss:\n", indent, ""); 118 segment_dump(stream, &sregs->ss, indent + 2); 119 fprintf(stream, "%*str:\n", indent, ""); 120 segment_dump(stream, &sregs->tr, indent + 2); 121 fprintf(stream, "%*sldt:\n", indent, ""); 122 segment_dump(stream, &sregs->ldt, indent + 2); 123 124 fprintf(stream, "%*sgdt:\n", indent, ""); 125 dtable_dump(stream, &sregs->gdt, indent + 2); 126 fprintf(stream, "%*sidt:\n", indent, ""); 127 dtable_dump(stream, &sregs->idt, indent + 2); 128 129 fprintf(stream, "%*scr0: 0x%.16llx cr2: 0x%.16llx " 130 "cr3: 0x%.16llx cr4: 0x%.16llx\n", 131 indent, "", 132 sregs->cr0, sregs->cr2, sregs->cr3, sregs->cr4); 133 fprintf(stream, "%*scr8: 0x%.16llx efer: 0x%.16llx " 134 "apic_base: 0x%.16llx\n", 135 indent, "", 136 sregs->cr8, sregs->efer, sregs->apic_base); 137 138 fprintf(stream, "%*sinterrupt_bitmap:\n", indent, ""); 139 for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++) { 140 fprintf(stream, "%*s%.16llx\n", indent + 2, "", 141 sregs->interrupt_bitmap[i]); 142 } 143 } 144 145 void virt_pgd_alloc(struct kvm_vm *vm) 146 { 147 TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, "Attempt to use " 148 "unknown or unsupported guest mode, mode: 0x%x", vm->mode); 149 150 /* If needed, create page map l4 table. */ 151 if (!vm->pgd_created) { 152 vm->pgd = vm_alloc_page_table(vm); 153 vm->pgd_created = true; 154 } 155 } 156 157 static void *virt_get_pte(struct kvm_vm *vm, uint64_t pt_pfn, uint64_t vaddr, 158 int level) 159 { 160 uint64_t *page_table = addr_gpa2hva(vm, pt_pfn << vm->page_shift); 161 int index = (vaddr >> PG_LEVEL_SHIFT(level)) & 0x1ffu; 162 163 return &page_table[index]; 164 } 165 166 static uint64_t *virt_create_upper_pte(struct kvm_vm *vm, 167 uint64_t pt_pfn, 168 uint64_t vaddr, 169 uint64_t paddr, 170 int current_level, 171 int target_level) 172 { 173 uint64_t *pte = virt_get_pte(vm, pt_pfn, vaddr, current_level); 174 175 if (!(*pte & PTE_PRESENT_MASK)) { 176 *pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK; 177 if (current_level == target_level) 178 *pte |= PTE_LARGE_MASK | (paddr & PHYSICAL_PAGE_MASK); 179 else 180 *pte |= vm_alloc_page_table(vm) & PHYSICAL_PAGE_MASK; 181 } else { 182 /* 183 * Entry already present. Assert that the caller doesn't want 184 * a hugepage at this level, and that there isn't a hugepage at 185 * this level. 186 */ 187 TEST_ASSERT(current_level != target_level, 188 "Cannot create hugepage at level: %u, vaddr: 0x%lx\n", 189 current_level, vaddr); 190 TEST_ASSERT(!(*pte & PTE_LARGE_MASK), 191 "Cannot create page table at level: %u, vaddr: 0x%lx\n", 192 current_level, vaddr); 193 } 194 return pte; 195 } 196 197 void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level) 198 { 199 const uint64_t pg_size = PG_LEVEL_SIZE(level); 200 uint64_t *pml4e, *pdpe, *pde; 201 uint64_t *pte; 202 203 TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, 204 "Unknown or unsupported guest mode, mode: 0x%x", vm->mode); 205 206 TEST_ASSERT((vaddr % pg_size) == 0, 207 "Virtual address not aligned,\n" 208 "vaddr: 0x%lx page size: 0x%lx", vaddr, pg_size); 209 TEST_ASSERT(sparsebit_is_set(vm->vpages_valid, (vaddr >> vm->page_shift)), 210 "Invalid virtual address, vaddr: 0x%lx", vaddr); 211 TEST_ASSERT((paddr % pg_size) == 0, 212 "Physical address not aligned,\n" 213 " paddr: 0x%lx page size: 0x%lx", paddr, pg_size); 214 TEST_ASSERT((paddr >> vm->page_shift) <= vm->max_gfn, 215 "Physical address beyond maximum supported,\n" 216 " paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x", 217 paddr, vm->max_gfn, vm->page_size); 218 219 /* 220 * Allocate upper level page tables, if not already present. Return 221 * early if a hugepage was created. 222 */ 223 pml4e = virt_create_upper_pte(vm, vm->pgd >> vm->page_shift, 224 vaddr, paddr, PG_LEVEL_512G, level); 225 if (*pml4e & PTE_LARGE_MASK) 226 return; 227 228 pdpe = virt_create_upper_pte(vm, PTE_GET_PFN(*pml4e), vaddr, paddr, PG_LEVEL_1G, level); 229 if (*pdpe & PTE_LARGE_MASK) 230 return; 231 232 pde = virt_create_upper_pte(vm, PTE_GET_PFN(*pdpe), vaddr, paddr, PG_LEVEL_2M, level); 233 if (*pde & PTE_LARGE_MASK) 234 return; 235 236 /* Fill in page table entry. */ 237 pte = virt_get_pte(vm, PTE_GET_PFN(*pde), vaddr, PG_LEVEL_4K); 238 TEST_ASSERT(!(*pte & PTE_PRESENT_MASK), 239 "PTE already present for 4k page at vaddr: 0x%lx\n", vaddr); 240 *pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK | (paddr & PHYSICAL_PAGE_MASK); 241 } 242 243 void virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr) 244 { 245 __virt_pg_map(vm, vaddr, paddr, PG_LEVEL_4K); 246 } 247 248 static uint64_t *_vm_get_page_table_entry(struct kvm_vm *vm, int vcpuid, 249 uint64_t vaddr) 250 { 251 uint16_t index[4]; 252 uint64_t *pml4e, *pdpe, *pde; 253 uint64_t *pte; 254 struct kvm_cpuid_entry2 *entry; 255 struct kvm_sregs sregs; 256 int max_phy_addr; 257 uint64_t rsvd_mask = 0; 258 259 entry = kvm_get_supported_cpuid_index(0x80000008, 0); 260 max_phy_addr = entry->eax & 0x000000ff; 261 /* Set the high bits in the reserved mask. */ 262 if (max_phy_addr < 52) 263 rsvd_mask = GENMASK_ULL(51, max_phy_addr); 264 265 /* 266 * SDM vol 3, fig 4-11 "Formats of CR3 and Paging-Structure Entries 267 * with 4-Level Paging and 5-Level Paging". 268 * If IA32_EFER.NXE = 0 and the P flag of a paging-structure entry is 1, 269 * the XD flag (bit 63) is reserved. 270 */ 271 vcpu_sregs_get(vm, vcpuid, &sregs); 272 if ((sregs.efer & EFER_NX) == 0) { 273 rsvd_mask |= PTE_NX_MASK; 274 } 275 276 TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, "Attempt to use " 277 "unknown or unsupported guest mode, mode: 0x%x", vm->mode); 278 TEST_ASSERT(sparsebit_is_set(vm->vpages_valid, 279 (vaddr >> vm->page_shift)), 280 "Invalid virtual address, vaddr: 0x%lx", 281 vaddr); 282 /* 283 * Based on the mode check above there are 48 bits in the vaddr, so 284 * shift 16 to sign extend the last bit (bit-47), 285 */ 286 TEST_ASSERT(vaddr == (((int64_t)vaddr << 16) >> 16), 287 "Canonical check failed. The virtual address is invalid."); 288 289 index[0] = (vaddr >> 12) & 0x1ffu; 290 index[1] = (vaddr >> 21) & 0x1ffu; 291 index[2] = (vaddr >> 30) & 0x1ffu; 292 index[3] = (vaddr >> 39) & 0x1ffu; 293 294 pml4e = addr_gpa2hva(vm, vm->pgd); 295 TEST_ASSERT(pml4e[index[3]] & PTE_PRESENT_MASK, 296 "Expected pml4e to be present for gva: 0x%08lx", vaddr); 297 TEST_ASSERT((pml4e[index[3]] & (rsvd_mask | PTE_LARGE_MASK)) == 0, 298 "Unexpected reserved bits set."); 299 300 pdpe = addr_gpa2hva(vm, PTE_GET_PFN(pml4e[index[3]]) * vm->page_size); 301 TEST_ASSERT(pdpe[index[2]] & PTE_PRESENT_MASK, 302 "Expected pdpe to be present for gva: 0x%08lx", vaddr); 303 TEST_ASSERT(!(pdpe[index[2]] & PTE_LARGE_MASK), 304 "Expected pdpe to map a pde not a 1-GByte page."); 305 TEST_ASSERT((pdpe[index[2]] & rsvd_mask) == 0, 306 "Unexpected reserved bits set."); 307 308 pde = addr_gpa2hva(vm, PTE_GET_PFN(pdpe[index[2]]) * vm->page_size); 309 TEST_ASSERT(pde[index[1]] & PTE_PRESENT_MASK, 310 "Expected pde to be present for gva: 0x%08lx", vaddr); 311 TEST_ASSERT(!(pde[index[1]] & PTE_LARGE_MASK), 312 "Expected pde to map a pte not a 2-MByte page."); 313 TEST_ASSERT((pde[index[1]] & rsvd_mask) == 0, 314 "Unexpected reserved bits set."); 315 316 pte = addr_gpa2hva(vm, PTE_GET_PFN(pde[index[1]]) * vm->page_size); 317 TEST_ASSERT(pte[index[0]] & PTE_PRESENT_MASK, 318 "Expected pte to be present for gva: 0x%08lx", vaddr); 319 320 return &pte[index[0]]; 321 } 322 323 uint64_t vm_get_page_table_entry(struct kvm_vm *vm, int vcpuid, uint64_t vaddr) 324 { 325 uint64_t *pte = _vm_get_page_table_entry(vm, vcpuid, vaddr); 326 327 return *(uint64_t *)pte; 328 } 329 330 void vm_set_page_table_entry(struct kvm_vm *vm, int vcpuid, uint64_t vaddr, 331 uint64_t pte) 332 { 333 uint64_t *new_pte = _vm_get_page_table_entry(vm, vcpuid, vaddr); 334 335 *(uint64_t *)new_pte = pte; 336 } 337 338 void virt_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent) 339 { 340 uint64_t *pml4e, *pml4e_start; 341 uint64_t *pdpe, *pdpe_start; 342 uint64_t *pde, *pde_start; 343 uint64_t *pte, *pte_start; 344 345 if (!vm->pgd_created) 346 return; 347 348 fprintf(stream, "%*s " 349 " no\n", indent, ""); 350 fprintf(stream, "%*s index hvaddr gpaddr " 351 "addr w exec dirty\n", 352 indent, ""); 353 pml4e_start = (uint64_t *) addr_gpa2hva(vm, vm->pgd); 354 for (uint16_t n1 = 0; n1 <= 0x1ffu; n1++) { 355 pml4e = &pml4e_start[n1]; 356 if (!(*pml4e & PTE_PRESENT_MASK)) 357 continue; 358 fprintf(stream, "%*spml4e 0x%-3zx %p 0x%-12lx 0x%-10llx %u " 359 " %u\n", 360 indent, "", 361 pml4e - pml4e_start, pml4e, 362 addr_hva2gpa(vm, pml4e), PTE_GET_PFN(*pml4e), 363 !!(*pml4e & PTE_WRITABLE_MASK), !!(*pml4e & PTE_NX_MASK)); 364 365 pdpe_start = addr_gpa2hva(vm, *pml4e & PHYSICAL_PAGE_MASK); 366 for (uint16_t n2 = 0; n2 <= 0x1ffu; n2++) { 367 pdpe = &pdpe_start[n2]; 368 if (!(*pdpe & PTE_PRESENT_MASK)) 369 continue; 370 fprintf(stream, "%*spdpe 0x%-3zx %p 0x%-12lx 0x%-10llx " 371 "%u %u\n", 372 indent, "", 373 pdpe - pdpe_start, pdpe, 374 addr_hva2gpa(vm, pdpe), 375 PTE_GET_PFN(*pdpe), !!(*pdpe & PTE_WRITABLE_MASK), 376 !!(*pdpe & PTE_NX_MASK)); 377 378 pde_start = addr_gpa2hva(vm, *pdpe & PHYSICAL_PAGE_MASK); 379 for (uint16_t n3 = 0; n3 <= 0x1ffu; n3++) { 380 pde = &pde_start[n3]; 381 if (!(*pde & PTE_PRESENT_MASK)) 382 continue; 383 fprintf(stream, "%*spde 0x%-3zx %p " 384 "0x%-12lx 0x%-10llx %u %u\n", 385 indent, "", pde - pde_start, pde, 386 addr_hva2gpa(vm, pde), 387 PTE_GET_PFN(*pde), !!(*pde & PTE_WRITABLE_MASK), 388 !!(*pde & PTE_NX_MASK)); 389 390 pte_start = addr_gpa2hva(vm, *pde & PHYSICAL_PAGE_MASK); 391 for (uint16_t n4 = 0; n4 <= 0x1ffu; n4++) { 392 pte = &pte_start[n4]; 393 if (!(*pte & PTE_PRESENT_MASK)) 394 continue; 395 fprintf(stream, "%*spte 0x%-3zx %p " 396 "0x%-12lx 0x%-10llx %u %u " 397 " %u 0x%-10lx\n", 398 indent, "", 399 pte - pte_start, pte, 400 addr_hva2gpa(vm, pte), 401 PTE_GET_PFN(*pte), 402 !!(*pte & PTE_WRITABLE_MASK), 403 !!(*pte & PTE_NX_MASK), 404 !!(*pte & PTE_DIRTY_MASK), 405 ((uint64_t) n1 << 27) 406 | ((uint64_t) n2 << 18) 407 | ((uint64_t) n3 << 9) 408 | ((uint64_t) n4)); 409 } 410 } 411 } 412 } 413 } 414 415 /* 416 * Set Unusable Segment 417 * 418 * Input Args: None 419 * 420 * Output Args: 421 * segp - Pointer to segment register 422 * 423 * Return: None 424 * 425 * Sets the segment register pointed to by @segp to an unusable state. 426 */ 427 static void kvm_seg_set_unusable(struct kvm_segment *segp) 428 { 429 memset(segp, 0, sizeof(*segp)); 430 segp->unusable = true; 431 } 432 433 static void kvm_seg_fill_gdt_64bit(struct kvm_vm *vm, struct kvm_segment *segp) 434 { 435 void *gdt = addr_gva2hva(vm, vm->gdt); 436 struct desc64 *desc = gdt + (segp->selector >> 3) * 8; 437 438 desc->limit0 = segp->limit & 0xFFFF; 439 desc->base0 = segp->base & 0xFFFF; 440 desc->base1 = segp->base >> 16; 441 desc->type = segp->type; 442 desc->s = segp->s; 443 desc->dpl = segp->dpl; 444 desc->p = segp->present; 445 desc->limit1 = segp->limit >> 16; 446 desc->avl = segp->avl; 447 desc->l = segp->l; 448 desc->db = segp->db; 449 desc->g = segp->g; 450 desc->base2 = segp->base >> 24; 451 if (!segp->s) 452 desc->base3 = segp->base >> 32; 453 } 454 455 456 /* 457 * Set Long Mode Flat Kernel Code Segment 458 * 459 * Input Args: 460 * vm - VM whose GDT is being filled, or NULL to only write segp 461 * selector - selector value 462 * 463 * Output Args: 464 * segp - Pointer to KVM segment 465 * 466 * Return: None 467 * 468 * Sets up the KVM segment pointed to by @segp, to be a code segment 469 * with the selector value given by @selector. 470 */ 471 static void kvm_seg_set_kernel_code_64bit(struct kvm_vm *vm, uint16_t selector, 472 struct kvm_segment *segp) 473 { 474 memset(segp, 0, sizeof(*segp)); 475 segp->selector = selector; 476 segp->limit = 0xFFFFFFFFu; 477 segp->s = 0x1; /* kTypeCodeData */ 478 segp->type = 0x08 | 0x01 | 0x02; /* kFlagCode | kFlagCodeAccessed 479 * | kFlagCodeReadable 480 */ 481 segp->g = true; 482 segp->l = true; 483 segp->present = 1; 484 if (vm) 485 kvm_seg_fill_gdt_64bit(vm, segp); 486 } 487 488 /* 489 * Set Long Mode Flat Kernel Data Segment 490 * 491 * Input Args: 492 * vm - VM whose GDT is being filled, or NULL to only write segp 493 * selector - selector value 494 * 495 * Output Args: 496 * segp - Pointer to KVM segment 497 * 498 * Return: None 499 * 500 * Sets up the KVM segment pointed to by @segp, to be a data segment 501 * with the selector value given by @selector. 502 */ 503 static void kvm_seg_set_kernel_data_64bit(struct kvm_vm *vm, uint16_t selector, 504 struct kvm_segment *segp) 505 { 506 memset(segp, 0, sizeof(*segp)); 507 segp->selector = selector; 508 segp->limit = 0xFFFFFFFFu; 509 segp->s = 0x1; /* kTypeCodeData */ 510 segp->type = 0x00 | 0x01 | 0x02; /* kFlagData | kFlagDataAccessed 511 * | kFlagDataWritable 512 */ 513 segp->g = true; 514 segp->present = true; 515 if (vm) 516 kvm_seg_fill_gdt_64bit(vm, segp); 517 } 518 519 vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva) 520 { 521 uint16_t index[4]; 522 uint64_t *pml4e, *pdpe, *pde; 523 uint64_t *pte; 524 525 TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, "Attempt to use " 526 "unknown or unsupported guest mode, mode: 0x%x", vm->mode); 527 528 index[0] = (gva >> 12) & 0x1ffu; 529 index[1] = (gva >> 21) & 0x1ffu; 530 index[2] = (gva >> 30) & 0x1ffu; 531 index[3] = (gva >> 39) & 0x1ffu; 532 533 if (!vm->pgd_created) 534 goto unmapped_gva; 535 pml4e = addr_gpa2hva(vm, vm->pgd); 536 if (!(pml4e[index[3]] & PTE_PRESENT_MASK)) 537 goto unmapped_gva; 538 539 pdpe = addr_gpa2hva(vm, PTE_GET_PFN(pml4e[index[3]]) * vm->page_size); 540 if (!(pdpe[index[2]] & PTE_PRESENT_MASK)) 541 goto unmapped_gva; 542 543 pde = addr_gpa2hva(vm, PTE_GET_PFN(pdpe[index[2]]) * vm->page_size); 544 if (!(pde[index[1]] & PTE_PRESENT_MASK)) 545 goto unmapped_gva; 546 547 pte = addr_gpa2hva(vm, PTE_GET_PFN(pde[index[1]]) * vm->page_size); 548 if (!(pte[index[0]] & PTE_PRESENT_MASK)) 549 goto unmapped_gva; 550 551 return (PTE_GET_PFN(pte[index[0]]) * vm->page_size) + (gva & ~PAGE_MASK); 552 553 unmapped_gva: 554 TEST_FAIL("No mapping for vm virtual address, gva: 0x%lx", gva); 555 exit(EXIT_FAILURE); 556 } 557 558 static void kvm_setup_gdt(struct kvm_vm *vm, struct kvm_dtable *dt) 559 { 560 if (!vm->gdt) 561 vm->gdt = vm_vaddr_alloc_page(vm); 562 563 dt->base = vm->gdt; 564 dt->limit = getpagesize(); 565 } 566 567 static void kvm_setup_tss_64bit(struct kvm_vm *vm, struct kvm_segment *segp, 568 int selector) 569 { 570 if (!vm->tss) 571 vm->tss = vm_vaddr_alloc_page(vm); 572 573 memset(segp, 0, sizeof(*segp)); 574 segp->base = vm->tss; 575 segp->limit = 0x67; 576 segp->selector = selector; 577 segp->type = 0xb; 578 segp->present = 1; 579 kvm_seg_fill_gdt_64bit(vm, segp); 580 } 581 582 static void vcpu_setup(struct kvm_vm *vm, int vcpuid) 583 { 584 struct kvm_sregs sregs; 585 586 /* Set mode specific system register values. */ 587 vcpu_sregs_get(vm, vcpuid, &sregs); 588 589 sregs.idt.limit = 0; 590 591 kvm_setup_gdt(vm, &sregs.gdt); 592 593 switch (vm->mode) { 594 case VM_MODE_PXXV48_4K: 595 sregs.cr0 = X86_CR0_PE | X86_CR0_NE | X86_CR0_PG; 596 sregs.cr4 |= X86_CR4_PAE | X86_CR4_OSFXSR; 597 sregs.efer |= (EFER_LME | EFER_LMA | EFER_NX); 598 599 kvm_seg_set_unusable(&sregs.ldt); 600 kvm_seg_set_kernel_code_64bit(vm, DEFAULT_CODE_SELECTOR, &sregs.cs); 601 kvm_seg_set_kernel_data_64bit(vm, DEFAULT_DATA_SELECTOR, &sregs.ds); 602 kvm_seg_set_kernel_data_64bit(vm, DEFAULT_DATA_SELECTOR, &sregs.es); 603 kvm_setup_tss_64bit(vm, &sregs.tr, 0x18); 604 break; 605 606 default: 607 TEST_FAIL("Unknown guest mode, mode: 0x%x", vm->mode); 608 } 609 610 sregs.cr3 = vm->pgd; 611 vcpu_sregs_set(vm, vcpuid, &sregs); 612 } 613 614 #define CPUID_XFD_BIT (1 << 4) 615 static bool is_xfd_supported(void) 616 { 617 int eax, ebx, ecx, edx; 618 const int leaf = 0xd, subleaf = 0x1; 619 620 __asm__ __volatile__( 621 "cpuid" 622 : /* output */ "=a"(eax), "=b"(ebx), 623 "=c"(ecx), "=d"(edx) 624 : /* input */ "0"(leaf), "2"(subleaf)); 625 626 return !!(eax & CPUID_XFD_BIT); 627 } 628 629 void vm_xsave_req_perm(int bit) 630 { 631 int kvm_fd; 632 u64 bitmask; 633 long rc; 634 struct kvm_device_attr attr = { 635 .group = 0, 636 .attr = KVM_X86_XCOMP_GUEST_SUPP, 637 .addr = (unsigned long) &bitmask 638 }; 639 640 kvm_fd = open_kvm_dev_path_or_exit(); 641 rc = ioctl(kvm_fd, KVM_GET_DEVICE_ATTR, &attr); 642 close(kvm_fd); 643 if (rc == -1 && (errno == ENXIO || errno == EINVAL)) 644 exit(KSFT_SKIP); 645 TEST_ASSERT(rc == 0, "KVM_GET_DEVICE_ATTR(0, KVM_X86_XCOMP_GUEST_SUPP) error: %ld", rc); 646 if (!(bitmask & (1ULL << bit))) 647 exit(KSFT_SKIP); 648 649 if (!is_xfd_supported()) 650 exit(KSFT_SKIP); 651 652 rc = syscall(SYS_arch_prctl, ARCH_REQ_XCOMP_GUEST_PERM, bit); 653 654 /* 655 * The older kernel version(<5.15) can't support 656 * ARCH_REQ_XCOMP_GUEST_PERM and directly return. 657 */ 658 if (rc) 659 return; 660 661 rc = syscall(SYS_arch_prctl, ARCH_GET_XCOMP_GUEST_PERM, &bitmask); 662 TEST_ASSERT(rc == 0, "prctl(ARCH_GET_XCOMP_GUEST_PERM) error: %ld", rc); 663 TEST_ASSERT(bitmask & (1ULL << bit), 664 "prctl(ARCH_REQ_XCOMP_GUEST_PERM) failure bitmask=0x%lx", 665 bitmask); 666 } 667 668 void vm_vcpu_add_default(struct kvm_vm *vm, uint32_t vcpuid, void *guest_code) 669 { 670 struct kvm_mp_state mp_state; 671 struct kvm_regs regs; 672 vm_vaddr_t stack_vaddr; 673 stack_vaddr = vm_vaddr_alloc(vm, DEFAULT_STACK_PGS * getpagesize(), 674 DEFAULT_GUEST_STACK_VADDR_MIN); 675 676 /* Create VCPU */ 677 vm_vcpu_add(vm, vcpuid); 678 vcpu_set_cpuid(vm, vcpuid, kvm_get_supported_cpuid()); 679 vcpu_setup(vm, vcpuid); 680 681 /* Setup guest general purpose registers */ 682 vcpu_regs_get(vm, vcpuid, ®s); 683 regs.rflags = regs.rflags | 0x2; 684 regs.rsp = stack_vaddr + (DEFAULT_STACK_PGS * getpagesize()); 685 regs.rip = (unsigned long) guest_code; 686 vcpu_regs_set(vm, vcpuid, ®s); 687 688 /* Setup the MP state */ 689 mp_state.mp_state = 0; 690 vcpu_set_mp_state(vm, vcpuid, &mp_state); 691 } 692 693 /* 694 * Allocate an instance of struct kvm_cpuid2 695 * 696 * Input Args: None 697 * 698 * Output Args: None 699 * 700 * Return: A pointer to the allocated struct. The caller is responsible 701 * for freeing this struct. 702 * 703 * Since kvm_cpuid2 uses a 0-length array to allow a the size of the 704 * array to be decided at allocation time, allocation is slightly 705 * complicated. This function uses a reasonable default length for 706 * the array and performs the appropriate allocation. 707 */ 708 static struct kvm_cpuid2 *allocate_kvm_cpuid2(void) 709 { 710 struct kvm_cpuid2 *cpuid; 711 int nent = 100; 712 size_t size; 713 714 size = sizeof(*cpuid); 715 size += nent * sizeof(struct kvm_cpuid_entry2); 716 cpuid = malloc(size); 717 if (!cpuid) { 718 perror("malloc"); 719 abort(); 720 } 721 722 cpuid->nent = nent; 723 724 return cpuid; 725 } 726 727 /* 728 * KVM Supported CPUID Get 729 * 730 * Input Args: None 731 * 732 * Output Args: 733 * 734 * Return: The supported KVM CPUID 735 * 736 * Get the guest CPUID supported by KVM. 737 */ 738 struct kvm_cpuid2 *kvm_get_supported_cpuid(void) 739 { 740 static struct kvm_cpuid2 *cpuid; 741 int ret; 742 int kvm_fd; 743 744 if (cpuid) 745 return cpuid; 746 747 cpuid = allocate_kvm_cpuid2(); 748 kvm_fd = open_kvm_dev_path_or_exit(); 749 750 ret = ioctl(kvm_fd, KVM_GET_SUPPORTED_CPUID, cpuid); 751 TEST_ASSERT(ret == 0, "KVM_GET_SUPPORTED_CPUID failed %d %d\n", 752 ret, errno); 753 754 close(kvm_fd); 755 return cpuid; 756 } 757 758 /* 759 * KVM Get MSR 760 * 761 * Input Args: 762 * msr_index - Index of MSR 763 * 764 * Output Args: None 765 * 766 * Return: On success, value of the MSR. On failure a TEST_ASSERT is produced. 767 * 768 * Get value of MSR for VCPU. 769 */ 770 uint64_t kvm_get_feature_msr(uint64_t msr_index) 771 { 772 struct { 773 struct kvm_msrs header; 774 struct kvm_msr_entry entry; 775 } buffer = {}; 776 int r, kvm_fd; 777 778 buffer.header.nmsrs = 1; 779 buffer.entry.index = msr_index; 780 kvm_fd = open_kvm_dev_path_or_exit(); 781 782 r = ioctl(kvm_fd, KVM_GET_MSRS, &buffer.header); 783 TEST_ASSERT(r == 1, "KVM_GET_MSRS IOCTL failed,\n" 784 " rc: %i errno: %i", r, errno); 785 786 close(kvm_fd); 787 return buffer.entry.data; 788 } 789 790 /* 791 * VM VCPU CPUID Set 792 * 793 * Input Args: 794 * vm - Virtual Machine 795 * vcpuid - VCPU id 796 * 797 * Output Args: None 798 * 799 * Return: KVM CPUID (KVM_GET_CPUID2) 800 * 801 * Set the VCPU's CPUID. 802 */ 803 struct kvm_cpuid2 *vcpu_get_cpuid(struct kvm_vm *vm, uint32_t vcpuid) 804 { 805 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 806 struct kvm_cpuid2 *cpuid; 807 int max_ent; 808 int rc = -1; 809 810 TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid); 811 812 cpuid = allocate_kvm_cpuid2(); 813 max_ent = cpuid->nent; 814 815 for (cpuid->nent = 1; cpuid->nent <= max_ent; cpuid->nent++) { 816 rc = ioctl(vcpu->fd, KVM_GET_CPUID2, cpuid); 817 if (!rc) 818 break; 819 820 TEST_ASSERT(rc == -1 && errno == E2BIG, 821 "KVM_GET_CPUID2 should either succeed or give E2BIG: %d %d", 822 rc, errno); 823 } 824 825 TEST_ASSERT(rc == 0, "KVM_GET_CPUID2 failed, rc: %i errno: %i", 826 rc, errno); 827 828 return cpuid; 829 } 830 831 832 833 /* 834 * Locate a cpuid entry. 835 * 836 * Input Args: 837 * function: The function of the cpuid entry to find. 838 * index: The index of the cpuid entry. 839 * 840 * Output Args: None 841 * 842 * Return: A pointer to the cpuid entry. Never returns NULL. 843 */ 844 struct kvm_cpuid_entry2 * 845 kvm_get_supported_cpuid_index(uint32_t function, uint32_t index) 846 { 847 struct kvm_cpuid2 *cpuid; 848 struct kvm_cpuid_entry2 *entry = NULL; 849 int i; 850 851 cpuid = kvm_get_supported_cpuid(); 852 for (i = 0; i < cpuid->nent; i++) { 853 if (cpuid->entries[i].function == function && 854 cpuid->entries[i].index == index) { 855 entry = &cpuid->entries[i]; 856 break; 857 } 858 } 859 860 TEST_ASSERT(entry, "Guest CPUID entry not found: (EAX=%x, ECX=%x).", 861 function, index); 862 return entry; 863 } 864 865 866 int __vcpu_set_cpuid(struct kvm_vm *vm, uint32_t vcpuid, 867 struct kvm_cpuid2 *cpuid) 868 { 869 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 870 871 TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid); 872 873 return ioctl(vcpu->fd, KVM_SET_CPUID2, cpuid); 874 } 875 876 /* 877 * VM VCPU CPUID Set 878 * 879 * Input Args: 880 * vm - Virtual Machine 881 * vcpuid - VCPU id 882 * cpuid - The CPUID values to set. 883 * 884 * Output Args: None 885 * 886 * Return: void 887 * 888 * Set the VCPU's CPUID. 889 */ 890 void vcpu_set_cpuid(struct kvm_vm *vm, 891 uint32_t vcpuid, struct kvm_cpuid2 *cpuid) 892 { 893 int rc; 894 895 rc = __vcpu_set_cpuid(vm, vcpuid, cpuid); 896 TEST_ASSERT(rc == 0, "KVM_SET_CPUID2 failed, rc: %i errno: %i", 897 rc, errno); 898 899 } 900 901 /* 902 * VCPU Get MSR 903 * 904 * Input Args: 905 * vm - Virtual Machine 906 * vcpuid - VCPU ID 907 * msr_index - Index of MSR 908 * 909 * Output Args: None 910 * 911 * Return: On success, value of the MSR. On failure a TEST_ASSERT is produced. 912 * 913 * Get value of MSR for VCPU. 914 */ 915 uint64_t vcpu_get_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index) 916 { 917 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 918 struct { 919 struct kvm_msrs header; 920 struct kvm_msr_entry entry; 921 } buffer = {}; 922 int r; 923 924 TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid); 925 buffer.header.nmsrs = 1; 926 buffer.entry.index = msr_index; 927 r = ioctl(vcpu->fd, KVM_GET_MSRS, &buffer.header); 928 TEST_ASSERT(r == 1, "KVM_GET_MSRS IOCTL failed,\n" 929 " rc: %i errno: %i", r, errno); 930 931 return buffer.entry.data; 932 } 933 934 /* 935 * _VCPU Set MSR 936 * 937 * Input Args: 938 * vm - Virtual Machine 939 * vcpuid - VCPU ID 940 * msr_index - Index of MSR 941 * msr_value - New value of MSR 942 * 943 * Output Args: None 944 * 945 * Return: The result of KVM_SET_MSRS. 946 * 947 * Sets the value of an MSR for the given VCPU. 948 */ 949 int _vcpu_set_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index, 950 uint64_t msr_value) 951 { 952 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 953 struct { 954 struct kvm_msrs header; 955 struct kvm_msr_entry entry; 956 } buffer = {}; 957 int r; 958 959 TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid); 960 memset(&buffer, 0, sizeof(buffer)); 961 buffer.header.nmsrs = 1; 962 buffer.entry.index = msr_index; 963 buffer.entry.data = msr_value; 964 r = ioctl(vcpu->fd, KVM_SET_MSRS, &buffer.header); 965 return r; 966 } 967 968 /* 969 * VCPU Set MSR 970 * 971 * Input Args: 972 * vm - Virtual Machine 973 * vcpuid - VCPU ID 974 * msr_index - Index of MSR 975 * msr_value - New value of MSR 976 * 977 * Output Args: None 978 * 979 * Return: On success, nothing. On failure a TEST_ASSERT is produced. 980 * 981 * Set value of MSR for VCPU. 982 */ 983 void vcpu_set_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index, 984 uint64_t msr_value) 985 { 986 int r; 987 988 r = _vcpu_set_msr(vm, vcpuid, msr_index, msr_value); 989 TEST_ASSERT(r == 1, "KVM_SET_MSRS IOCTL failed,\n" 990 " rc: %i errno: %i", r, errno); 991 } 992 993 void vcpu_args_set(struct kvm_vm *vm, uint32_t vcpuid, unsigned int num, ...) 994 { 995 va_list ap; 996 struct kvm_regs regs; 997 998 TEST_ASSERT(num >= 1 && num <= 6, "Unsupported number of args,\n" 999 " num: %u\n", 1000 num); 1001 1002 va_start(ap, num); 1003 vcpu_regs_get(vm, vcpuid, ®s); 1004 1005 if (num >= 1) 1006 regs.rdi = va_arg(ap, uint64_t); 1007 1008 if (num >= 2) 1009 regs.rsi = va_arg(ap, uint64_t); 1010 1011 if (num >= 3) 1012 regs.rdx = va_arg(ap, uint64_t); 1013 1014 if (num >= 4) 1015 regs.rcx = va_arg(ap, uint64_t); 1016 1017 if (num >= 5) 1018 regs.r8 = va_arg(ap, uint64_t); 1019 1020 if (num >= 6) 1021 regs.r9 = va_arg(ap, uint64_t); 1022 1023 vcpu_regs_set(vm, vcpuid, ®s); 1024 va_end(ap); 1025 } 1026 1027 void vcpu_dump(FILE *stream, struct kvm_vm *vm, uint32_t vcpuid, uint8_t indent) 1028 { 1029 struct kvm_regs regs; 1030 struct kvm_sregs sregs; 1031 1032 fprintf(stream, "%*scpuid: %u\n", indent, "", vcpuid); 1033 1034 fprintf(stream, "%*sregs:\n", indent + 2, ""); 1035 vcpu_regs_get(vm, vcpuid, ®s); 1036 regs_dump(stream, ®s, indent + 4); 1037 1038 fprintf(stream, "%*ssregs:\n", indent + 2, ""); 1039 vcpu_sregs_get(vm, vcpuid, &sregs); 1040 sregs_dump(stream, &sregs, indent + 4); 1041 } 1042 1043 static int kvm_get_num_msrs_fd(int kvm_fd) 1044 { 1045 struct kvm_msr_list nmsrs; 1046 int r; 1047 1048 nmsrs.nmsrs = 0; 1049 r = ioctl(kvm_fd, KVM_GET_MSR_INDEX_LIST, &nmsrs); 1050 TEST_ASSERT(r == -1 && errno == E2BIG, "Unexpected result from KVM_GET_MSR_INDEX_LIST probe, r: %i", 1051 r); 1052 1053 return nmsrs.nmsrs; 1054 } 1055 1056 static int kvm_get_num_msrs(struct kvm_vm *vm) 1057 { 1058 return kvm_get_num_msrs_fd(vm->kvm_fd); 1059 } 1060 1061 struct kvm_msr_list *kvm_get_msr_index_list(void) 1062 { 1063 struct kvm_msr_list *list; 1064 int nmsrs, r, kvm_fd; 1065 1066 kvm_fd = open_kvm_dev_path_or_exit(); 1067 1068 nmsrs = kvm_get_num_msrs_fd(kvm_fd); 1069 list = malloc(sizeof(*list) + nmsrs * sizeof(list->indices[0])); 1070 list->nmsrs = nmsrs; 1071 r = ioctl(kvm_fd, KVM_GET_MSR_INDEX_LIST, list); 1072 close(kvm_fd); 1073 1074 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MSR_INDEX_LIST, r: %i", 1075 r); 1076 1077 return list; 1078 } 1079 1080 static int vcpu_save_xsave_state(struct kvm_vm *vm, struct vcpu *vcpu, 1081 struct kvm_x86_state *state) 1082 { 1083 int size; 1084 1085 size = vm_check_cap(vm, KVM_CAP_XSAVE2); 1086 if (!size) 1087 size = sizeof(struct kvm_xsave); 1088 1089 state->xsave = malloc(size); 1090 if (size == sizeof(struct kvm_xsave)) 1091 return ioctl(vcpu->fd, KVM_GET_XSAVE, state->xsave); 1092 else 1093 return ioctl(vcpu->fd, KVM_GET_XSAVE2, state->xsave); 1094 } 1095 1096 struct kvm_x86_state *vcpu_save_state(struct kvm_vm *vm, uint32_t vcpuid) 1097 { 1098 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 1099 struct kvm_msr_list *list; 1100 struct kvm_x86_state *state; 1101 int nmsrs, r, i; 1102 static int nested_size = -1; 1103 1104 if (nested_size == -1) { 1105 nested_size = kvm_check_cap(KVM_CAP_NESTED_STATE); 1106 TEST_ASSERT(nested_size <= sizeof(state->nested_), 1107 "Nested state size too big, %i > %zi", 1108 nested_size, sizeof(state->nested_)); 1109 } 1110 1111 /* 1112 * When KVM exits to userspace with KVM_EXIT_IO, KVM guarantees 1113 * guest state is consistent only after userspace re-enters the 1114 * kernel with KVM_RUN. Complete IO prior to migrating state 1115 * to a new VM. 1116 */ 1117 vcpu_run_complete_io(vm, vcpuid); 1118 1119 nmsrs = kvm_get_num_msrs(vm); 1120 list = malloc(sizeof(*list) + nmsrs * sizeof(list->indices[0])); 1121 list->nmsrs = nmsrs; 1122 r = ioctl(vm->kvm_fd, KVM_GET_MSR_INDEX_LIST, list); 1123 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MSR_INDEX_LIST, r: %i", 1124 r); 1125 1126 state = malloc(sizeof(*state) + nmsrs * sizeof(state->msrs.entries[0])); 1127 r = ioctl(vcpu->fd, KVM_GET_VCPU_EVENTS, &state->events); 1128 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_VCPU_EVENTS, r: %i", 1129 r); 1130 1131 r = ioctl(vcpu->fd, KVM_GET_MP_STATE, &state->mp_state); 1132 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MP_STATE, r: %i", 1133 r); 1134 1135 r = ioctl(vcpu->fd, KVM_GET_REGS, &state->regs); 1136 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_REGS, r: %i", 1137 r); 1138 1139 r = vcpu_save_xsave_state(vm, vcpu, state); 1140 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_XSAVE, r: %i", 1141 r); 1142 1143 if (kvm_check_cap(KVM_CAP_XCRS)) { 1144 r = ioctl(vcpu->fd, KVM_GET_XCRS, &state->xcrs); 1145 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_XCRS, r: %i", 1146 r); 1147 } 1148 1149 r = ioctl(vcpu->fd, KVM_GET_SREGS, &state->sregs); 1150 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_SREGS, r: %i", 1151 r); 1152 1153 if (nested_size) { 1154 state->nested.size = sizeof(state->nested_); 1155 r = ioctl(vcpu->fd, KVM_GET_NESTED_STATE, &state->nested); 1156 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_NESTED_STATE, r: %i", 1157 r); 1158 TEST_ASSERT(state->nested.size <= nested_size, 1159 "Nested state size too big, %i (KVM_CHECK_CAP gave %i)", 1160 state->nested.size, nested_size); 1161 } else 1162 state->nested.size = 0; 1163 1164 state->msrs.nmsrs = nmsrs; 1165 for (i = 0; i < nmsrs; i++) 1166 state->msrs.entries[i].index = list->indices[i]; 1167 r = ioctl(vcpu->fd, KVM_GET_MSRS, &state->msrs); 1168 TEST_ASSERT(r == nmsrs, "Unexpected result from KVM_GET_MSRS, r: %i (failed MSR was 0x%x)", 1169 r, r == nmsrs ? -1 : list->indices[r]); 1170 1171 r = ioctl(vcpu->fd, KVM_GET_DEBUGREGS, &state->debugregs); 1172 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_DEBUGREGS, r: %i", 1173 r); 1174 1175 free(list); 1176 return state; 1177 } 1178 1179 void vcpu_load_state(struct kvm_vm *vm, uint32_t vcpuid, struct kvm_x86_state *state) 1180 { 1181 struct vcpu *vcpu = vcpu_find(vm, vcpuid); 1182 int r; 1183 1184 r = ioctl(vcpu->fd, KVM_SET_SREGS, &state->sregs); 1185 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_SREGS, r: %i", 1186 r); 1187 1188 r = ioctl(vcpu->fd, KVM_SET_MSRS, &state->msrs); 1189 TEST_ASSERT(r == state->msrs.nmsrs, 1190 "Unexpected result from KVM_SET_MSRS, r: %i (failed at %x)", 1191 r, r == state->msrs.nmsrs ? -1 : state->msrs.entries[r].index); 1192 1193 if (kvm_check_cap(KVM_CAP_XCRS)) { 1194 r = ioctl(vcpu->fd, KVM_SET_XCRS, &state->xcrs); 1195 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_XCRS, r: %i", 1196 r); 1197 } 1198 1199 r = ioctl(vcpu->fd, KVM_SET_XSAVE, state->xsave); 1200 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_XSAVE, r: %i", 1201 r); 1202 1203 r = ioctl(vcpu->fd, KVM_SET_VCPU_EVENTS, &state->events); 1204 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_VCPU_EVENTS, r: %i", 1205 r); 1206 1207 r = ioctl(vcpu->fd, KVM_SET_MP_STATE, &state->mp_state); 1208 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_MP_STATE, r: %i", 1209 r); 1210 1211 r = ioctl(vcpu->fd, KVM_SET_DEBUGREGS, &state->debugregs); 1212 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_DEBUGREGS, r: %i", 1213 r); 1214 1215 r = ioctl(vcpu->fd, KVM_SET_REGS, &state->regs); 1216 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_REGS, r: %i", 1217 r); 1218 1219 if (state->nested.size) { 1220 r = ioctl(vcpu->fd, KVM_SET_NESTED_STATE, &state->nested); 1221 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_NESTED_STATE, r: %i", 1222 r); 1223 } 1224 } 1225 1226 void kvm_x86_state_cleanup(struct kvm_x86_state *state) 1227 { 1228 free(state->xsave); 1229 free(state); 1230 } 1231 1232 static bool cpu_vendor_string_is(const char *vendor) 1233 { 1234 const uint32_t *chunk = (const uint32_t *)vendor; 1235 int eax, ebx, ecx, edx; 1236 const int leaf = 0; 1237 1238 __asm__ __volatile__( 1239 "cpuid" 1240 : /* output */ "=a"(eax), "=b"(ebx), 1241 "=c"(ecx), "=d"(edx) 1242 : /* input */ "0"(leaf), "2"(0)); 1243 1244 return (ebx == chunk[0] && edx == chunk[1] && ecx == chunk[2]); 1245 } 1246 1247 bool is_intel_cpu(void) 1248 { 1249 return cpu_vendor_string_is("GenuineIntel"); 1250 } 1251 1252 /* 1253 * Exclude early K5 samples with a vendor string of "AMDisbetter!" 1254 */ 1255 bool is_amd_cpu(void) 1256 { 1257 return cpu_vendor_string_is("AuthenticAMD"); 1258 } 1259 1260 uint32_t kvm_get_cpuid_max_basic(void) 1261 { 1262 return kvm_get_supported_cpuid_entry(0)->eax; 1263 } 1264 1265 uint32_t kvm_get_cpuid_max_extended(void) 1266 { 1267 return kvm_get_supported_cpuid_entry(0x80000000)->eax; 1268 } 1269 1270 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits) 1271 { 1272 struct kvm_cpuid_entry2 *entry; 1273 bool pae; 1274 1275 /* SDM 4.1.4 */ 1276 if (kvm_get_cpuid_max_extended() < 0x80000008) { 1277 pae = kvm_get_supported_cpuid_entry(1)->edx & (1 << 6); 1278 *pa_bits = pae ? 36 : 32; 1279 *va_bits = 32; 1280 } else { 1281 entry = kvm_get_supported_cpuid_entry(0x80000008); 1282 *pa_bits = entry->eax & 0xff; 1283 *va_bits = (entry->eax >> 8) & 0xff; 1284 } 1285 } 1286 1287 struct idt_entry { 1288 uint16_t offset0; 1289 uint16_t selector; 1290 uint16_t ist : 3; 1291 uint16_t : 5; 1292 uint16_t type : 4; 1293 uint16_t : 1; 1294 uint16_t dpl : 2; 1295 uint16_t p : 1; 1296 uint16_t offset1; 1297 uint32_t offset2; uint32_t reserved; 1298 }; 1299 1300 static void set_idt_entry(struct kvm_vm *vm, int vector, unsigned long addr, 1301 int dpl, unsigned short selector) 1302 { 1303 struct idt_entry *base = 1304 (struct idt_entry *)addr_gva2hva(vm, vm->idt); 1305 struct idt_entry *e = &base[vector]; 1306 1307 memset(e, 0, sizeof(*e)); 1308 e->offset0 = addr; 1309 e->selector = selector; 1310 e->ist = 0; 1311 e->type = 14; 1312 e->dpl = dpl; 1313 e->p = 1; 1314 e->offset1 = addr >> 16; 1315 e->offset2 = addr >> 32; 1316 } 1317 1318 void kvm_exit_unexpected_vector(uint32_t value) 1319 { 1320 ucall(UCALL_UNHANDLED, 1, value); 1321 } 1322 1323 void route_exception(struct ex_regs *regs) 1324 { 1325 typedef void(*handler)(struct ex_regs *); 1326 handler *handlers = (handler *)exception_handlers; 1327 1328 if (handlers && handlers[regs->vector]) { 1329 handlers[regs->vector](regs); 1330 return; 1331 } 1332 1333 kvm_exit_unexpected_vector(regs->vector); 1334 } 1335 1336 void vm_init_descriptor_tables(struct kvm_vm *vm) 1337 { 1338 extern void *idt_handlers; 1339 int i; 1340 1341 vm->idt = vm_vaddr_alloc_page(vm); 1342 vm->handlers = vm_vaddr_alloc_page(vm); 1343 /* Handlers have the same address in both address spaces.*/ 1344 for (i = 0; i < NUM_INTERRUPTS; i++) 1345 set_idt_entry(vm, i, (unsigned long)(&idt_handlers)[i], 0, 1346 DEFAULT_CODE_SELECTOR); 1347 } 1348 1349 void vcpu_init_descriptor_tables(struct kvm_vm *vm, uint32_t vcpuid) 1350 { 1351 struct kvm_sregs sregs; 1352 1353 vcpu_sregs_get(vm, vcpuid, &sregs); 1354 sregs.idt.base = vm->idt; 1355 sregs.idt.limit = NUM_INTERRUPTS * sizeof(struct idt_entry) - 1; 1356 sregs.gdt.base = vm->gdt; 1357 sregs.gdt.limit = getpagesize() - 1; 1358 kvm_seg_set_kernel_data_64bit(NULL, DEFAULT_DATA_SELECTOR, &sregs.gs); 1359 vcpu_sregs_set(vm, vcpuid, &sregs); 1360 *(vm_vaddr_t *)addr_gva2hva(vm, (vm_vaddr_t)(&exception_handlers)) = vm->handlers; 1361 } 1362 1363 void vm_install_exception_handler(struct kvm_vm *vm, int vector, 1364 void (*handler)(struct ex_regs *)) 1365 { 1366 vm_vaddr_t *handlers = (vm_vaddr_t *)addr_gva2hva(vm, vm->handlers); 1367 1368 handlers[vector] = (vm_vaddr_t)handler; 1369 } 1370 1371 void assert_on_unhandled_exception(struct kvm_vm *vm, uint32_t vcpuid) 1372 { 1373 struct ucall uc; 1374 1375 if (get_ucall(vm, vcpuid, &uc) == UCALL_UNHANDLED) { 1376 uint64_t vector = uc.args[0]; 1377 1378 TEST_FAIL("Unexpected vectored event in guest (vector:0x%lx)", 1379 vector); 1380 } 1381 } 1382 1383 struct kvm_cpuid_entry2 *get_cpuid(struct kvm_cpuid2 *cpuid, uint32_t function, 1384 uint32_t index) 1385 { 1386 int i; 1387 1388 for (i = 0; i < cpuid->nent; i++) { 1389 struct kvm_cpuid_entry2 *cur = &cpuid->entries[i]; 1390 1391 if (cur->function == function && cur->index == index) 1392 return cur; 1393 } 1394 1395 TEST_FAIL("CPUID function 0x%x index 0x%x not found ", function, index); 1396 1397 return NULL; 1398 } 1399 1400 bool set_cpuid(struct kvm_cpuid2 *cpuid, 1401 struct kvm_cpuid_entry2 *ent) 1402 { 1403 int i; 1404 1405 for (i = 0; i < cpuid->nent; i++) { 1406 struct kvm_cpuid_entry2 *cur = &cpuid->entries[i]; 1407 1408 if (cur->function != ent->function || cur->index != ent->index) 1409 continue; 1410 1411 memcpy(cur, ent, sizeof(struct kvm_cpuid_entry2)); 1412 return true; 1413 } 1414 1415 return false; 1416 } 1417 1418 uint64_t kvm_hypercall(uint64_t nr, uint64_t a0, uint64_t a1, uint64_t a2, 1419 uint64_t a3) 1420 { 1421 uint64_t r; 1422 1423 asm volatile("vmcall" 1424 : "=a"(r) 1425 : "b"(a0), "c"(a1), "d"(a2), "S"(a3)); 1426 return r; 1427 } 1428 1429 struct kvm_cpuid2 *kvm_get_supported_hv_cpuid(void) 1430 { 1431 static struct kvm_cpuid2 *cpuid; 1432 int ret; 1433 int kvm_fd; 1434 1435 if (cpuid) 1436 return cpuid; 1437 1438 cpuid = allocate_kvm_cpuid2(); 1439 kvm_fd = open_kvm_dev_path_or_exit(); 1440 1441 ret = ioctl(kvm_fd, KVM_GET_SUPPORTED_HV_CPUID, cpuid); 1442 TEST_ASSERT(ret == 0, "KVM_GET_SUPPORTED_HV_CPUID failed %d %d\n", 1443 ret, errno); 1444 1445 close(kvm_fd); 1446 return cpuid; 1447 } 1448 1449 void vcpu_set_hv_cpuid(struct kvm_vm *vm, uint32_t vcpuid) 1450 { 1451 static struct kvm_cpuid2 *cpuid_full; 1452 struct kvm_cpuid2 *cpuid_sys, *cpuid_hv; 1453 int i, nent = 0; 1454 1455 if (!cpuid_full) { 1456 cpuid_sys = kvm_get_supported_cpuid(); 1457 cpuid_hv = kvm_get_supported_hv_cpuid(); 1458 1459 cpuid_full = malloc(sizeof(*cpuid_full) + 1460 (cpuid_sys->nent + cpuid_hv->nent) * 1461 sizeof(struct kvm_cpuid_entry2)); 1462 if (!cpuid_full) { 1463 perror("malloc"); 1464 abort(); 1465 } 1466 1467 /* Need to skip KVM CPUID leaves 0x400000xx */ 1468 for (i = 0; i < cpuid_sys->nent; i++) { 1469 if (cpuid_sys->entries[i].function >= 0x40000000 && 1470 cpuid_sys->entries[i].function < 0x40000100) 1471 continue; 1472 cpuid_full->entries[nent] = cpuid_sys->entries[i]; 1473 nent++; 1474 } 1475 1476 memcpy(&cpuid_full->entries[nent], cpuid_hv->entries, 1477 cpuid_hv->nent * sizeof(struct kvm_cpuid_entry2)); 1478 cpuid_full->nent = nent + cpuid_hv->nent; 1479 } 1480 1481 vcpu_set_cpuid(vm, vcpuid, cpuid_full); 1482 } 1483 1484 struct kvm_cpuid2 *vcpu_get_supported_hv_cpuid(struct kvm_vm *vm, uint32_t vcpuid) 1485 { 1486 static struct kvm_cpuid2 *cpuid; 1487 1488 cpuid = allocate_kvm_cpuid2(); 1489 1490 vcpu_ioctl(vm, vcpuid, KVM_GET_SUPPORTED_HV_CPUID, cpuid); 1491 1492 return cpuid; 1493 } 1494 1495 unsigned long vm_compute_max_gfn(struct kvm_vm *vm) 1496 { 1497 const unsigned long num_ht_pages = 12 << (30 - vm->page_shift); /* 12 GiB */ 1498 unsigned long ht_gfn, max_gfn, max_pfn; 1499 uint32_t eax, ebx, ecx, edx, max_ext_leaf; 1500 1501 max_gfn = (1ULL << (vm->pa_bits - vm->page_shift)) - 1; 1502 1503 /* Avoid reserved HyperTransport region on AMD processors. */ 1504 if (!is_amd_cpu()) 1505 return max_gfn; 1506 1507 /* On parts with <40 physical address bits, the area is fully hidden */ 1508 if (vm->pa_bits < 40) 1509 return max_gfn; 1510 1511 /* Before family 17h, the HyperTransport area is just below 1T. */ 1512 ht_gfn = (1 << 28) - num_ht_pages; 1513 eax = 1; 1514 ecx = 0; 1515 cpuid(&eax, &ebx, &ecx, &edx); 1516 if (x86_family(eax) < 0x17) 1517 goto done; 1518 1519 /* 1520 * Otherwise it's at the top of the physical address space, possibly 1521 * reduced due to SME by bits 11:6 of CPUID[0x8000001f].EBX. Use 1522 * the old conservative value if MAXPHYADDR is not enumerated. 1523 */ 1524 eax = 0x80000000; 1525 cpuid(&eax, &ebx, &ecx, &edx); 1526 max_ext_leaf = eax; 1527 if (max_ext_leaf < 0x80000008) 1528 goto done; 1529 1530 eax = 0x80000008; 1531 cpuid(&eax, &ebx, &ecx, &edx); 1532 max_pfn = (1ULL << ((eax & 0xff) - vm->page_shift)) - 1; 1533 if (max_ext_leaf >= 0x8000001f) { 1534 eax = 0x8000001f; 1535 cpuid(&eax, &ebx, &ecx, &edx); 1536 max_pfn >>= (ebx >> 6) & 0x3f; 1537 } 1538 1539 ht_gfn = max_pfn - num_ht_pages; 1540 done: 1541 return min(max_gfn, ht_gfn - 1); 1542 } 1543