tools/testing/selftests/kvm/lib/x86_64/processor.c

   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 #define CPUID_XFD_BIT (1 << 4)
 654 static bool is_xfd_supported(void)
 655 {
 656         int eax, ebx, ecx, edx;
 657         const int leaf = 0xd, subleaf = 0x1;
 658
 659         __asm__ __volatile__(
 660                 "cpuid"
 661                 : /* output */ "=a"(eax), "=b"(ebx),
 662                   "=c"(ecx), "=d"(edx)
 663                 : /* input */ "0"(leaf), "2"(subleaf));
 664
 665         return !!(eax & CPUID_XFD_BIT);
 666 }
 667
 668 void vm_xsave_req_perm(void)
 669 {
 670         unsigned long bitmask;
 671         long rc;
 672
 673         if (!is_xfd_supported())
 674                 return;
 675
 676         rc = syscall(SYS_arch_prctl, ARCH_REQ_XCOMP_GUEST_PERM,
 677                      XSTATE_XTILE_DATA_BIT);
 678         /*
 679          * The older kernel version(<5.15) can't support
 680          * ARCH_REQ_XCOMP_GUEST_PERM and directly return.
 681          */
 682         if (rc)
 683                 return;
 684
 685         rc = syscall(SYS_arch_prctl, ARCH_GET_XCOMP_GUEST_PERM, &bitmask);
 686         TEST_ASSERT(rc == 0, "prctl(ARCH_GET_XCOMP_GUEST_PERM) error: %ld", rc);
 687         TEST_ASSERT(bitmask & XFEATURE_XTILE_MASK,
 688                     "prctl(ARCH_REQ_XCOMP_GUEST_PERM) failure bitmask=0x%lx",
 689                     bitmask);
 690 }
 691
 692 void vm_vcpu_add_default(struct kvm_vm *vm, uint32_t vcpuid, void *guest_code)
 693 {
 694         struct kvm_mp_state mp_state;
 695         struct kvm_regs regs;
 696         vm_vaddr_t stack_vaddr;
 697         stack_vaddr = vm_vaddr_alloc(vm, DEFAULT_STACK_PGS * getpagesize(),
 698                                      DEFAULT_GUEST_STACK_VADDR_MIN);
 699
 700         /* Create VCPU */
 701         vm_vcpu_add(vm, vcpuid);
 702         vcpu_set_cpuid(vm, vcpuid, kvm_get_supported_cpuid());
 703         vcpu_setup(vm, vcpuid);
 704
 705         /* Setup guest general purpose registers */
 706         vcpu_regs_get(vm, vcpuid, &regs);
 707         regs.rflags = regs.rflags | 0x2;
 708         regs.rsp = stack_vaddr + (DEFAULT_STACK_PGS * getpagesize());
 709         regs.rip = (unsigned long) guest_code;
 710         vcpu_regs_set(vm, vcpuid, &regs);
 711
 712         /* Setup the MP state */
 713         mp_state.mp_state = 0;
 714         vcpu_set_mp_state(vm, vcpuid, &mp_state);
 715 }
 716
 717 /*
 718  * Allocate an instance of struct kvm_cpuid2
 719  *
 720  * Input Args: None
 721  *
 722  * Output Args: None
 723  *
 724  * Return: A pointer to the allocated struct. The caller is responsible
 725  * for freeing this struct.
 726  *
 727  * Since kvm_cpuid2 uses a 0-length array to allow a the size of the
 728  * array to be decided at allocation time, allocation is slightly
 729  * complicated. This function uses a reasonable default length for
 730  * the array and performs the appropriate allocation.
 731  */
 732 static struct kvm_cpuid2 *allocate_kvm_cpuid2(void)
 733 {
 734         struct kvm_cpuid2 *cpuid;
 735         int nent = 100;
 736         size_t size;
 737
 738         size = sizeof(*cpuid);
 739         size += nent * sizeof(struct kvm_cpuid_entry2);
 740         cpuid = malloc(size);
 741         if (!cpuid) {
 742                 perror("malloc");
 743                 abort();
 744         }
 745
 746         cpuid->nent = nent;
 747
 748         return cpuid;
 749 }
 750
 751 /*
 752  * KVM Supported CPUID Get
 753  *
 754  * Input Args: None
 755  *
 756  * Output Args:
 757  *
 758  * Return: The supported KVM CPUID
 759  *
 760  * Get the guest CPUID supported by KVM.
 761  */
 762 struct kvm_cpuid2 *kvm_get_supported_cpuid(void)
 763 {
 764         static struct kvm_cpuid2 *cpuid;
 765         int ret;
 766         int kvm_fd;
 767
 768         if (cpuid)
 769                 return cpuid;
 770
 771         cpuid = allocate_kvm_cpuid2();
 772         kvm_fd = open_kvm_dev_path_or_exit();
 773
 774         ret = ioctl(kvm_fd, KVM_GET_SUPPORTED_CPUID, cpuid);
 775         TEST_ASSERT(ret == 0, "KVM_GET_SUPPORTED_CPUID failed %d %d\n",
 776                     ret, errno);
 777
 778         close(kvm_fd);
 779         return cpuid;
 780 }
 781
 782 /*
 783  * KVM Get MSR
 784  *
 785  * Input Args:
 786  *   msr_index - Index of MSR
 787  *
 788  * Output Args: None
 789  *
 790  * Return: On success, value of the MSR. On failure a TEST_ASSERT is produced.
 791  *
 792  * Get value of MSR for VCPU.
 793  */
 794 uint64_t kvm_get_feature_msr(uint64_t msr_index)
 795 {
 796         struct {
 797                 struct kvm_msrs header;
 798                 struct kvm_msr_entry entry;
 799         } buffer = {};
 800         int r, kvm_fd;
 801
 802         buffer.header.nmsrs = 1;
 803         buffer.entry.index = msr_index;
 804         kvm_fd = open_kvm_dev_path_or_exit();
 805
 806         r = ioctl(kvm_fd, KVM_GET_MSRS, &buffer.header);
 807         TEST_ASSERT(r == 1, "KVM_GET_MSRS IOCTL failed,\n"
 808                 "  rc: %i errno: %i", r, errno);
 809
 810         close(kvm_fd);
 811         return buffer.entry.data;
 812 }
 813
 814 /*
 815  * VM VCPU CPUID Set
 816  *
 817  * Input Args:
 818  *   vm - Virtual Machine
 819  *   vcpuid - VCPU id
 820  *
 821  * Output Args: None
 822  *
 823  * Return: KVM CPUID (KVM_GET_CPUID2)
 824  *
 825  * Set the VCPU's CPUID.
 826  */
 827 struct kvm_cpuid2 *vcpu_get_cpuid(struct kvm_vm *vm, uint32_t vcpuid)
 828 {
 829         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
 830         struct kvm_cpuid2 *cpuid;
 831         int max_ent;
 832         int rc = -1;
 833
 834         TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid);
 835
 836         cpuid = allocate_kvm_cpuid2();
 837         max_ent = cpuid->nent;
 838
 839         for (cpuid->nent = 1; cpuid->nent <= max_ent; cpuid->nent++) {
 840                 rc = ioctl(vcpu->fd, KVM_GET_CPUID2, cpuid);
 841                 if (!rc)
 842                         break;
 843
 844                 TEST_ASSERT(rc == -1 && errno == E2BIG,
 845                             "KVM_GET_CPUID2 should either succeed or give E2BIG: %d %d",
 846                             rc, errno);
 847         }
 848
 849         TEST_ASSERT(rc == 0, "KVM_GET_CPUID2 failed, rc: %i errno: %i",
 850                     rc, errno);
 851
 852         return cpuid;
 853 }
 854
 855
 856
 857 /*
 858  * Locate a cpuid entry.
 859  *
 860  * Input Args:
 861  *   function: The function of the cpuid entry to find.
 862  *   index: The index of the cpuid entry.
 863  *
 864  * Output Args: None
 865  *
 866  * Return: A pointer to the cpuid entry. Never returns NULL.
 867  */
 868 struct kvm_cpuid_entry2 *
 869 kvm_get_supported_cpuid_index(uint32_t function, uint32_t index)
 870 {
 871         struct kvm_cpuid2 *cpuid;
 872         struct kvm_cpuid_entry2 *entry = NULL;
 873         int i;
 874
 875         cpuid = kvm_get_supported_cpuid();
 876         for (i = 0; i < cpuid->nent; i++) {
 877                 if (cpuid->entries[i].function == function &&
 878                     cpuid->entries[i].index == index) {
 879                         entry = &cpuid->entries[i];
 880                         break;
 881                 }
 882         }
 883
 884         TEST_ASSERT(entry, "Guest CPUID entry not found: (EAX=%x, ECX=%x).",
 885                     function, index);
 886         return entry;
 887 }
 888
 889
 890 int __vcpu_set_cpuid(struct kvm_vm *vm, uint32_t vcpuid,
 891                      struct kvm_cpuid2 *cpuid)
 892 {
 893         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
 894
 895         TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid);
 896
 897         return ioctl(vcpu->fd, KVM_SET_CPUID2, cpuid);
 898 }
 899
 900 /*
 901  * VM VCPU CPUID Set
 902  *
 903  * Input Args:
 904  *   vm - Virtual Machine
 905  *   vcpuid - VCPU id
 906  *   cpuid - The CPUID values to set.
 907  *
 908  * Output Args: None
 909  *
 910  * Return: void
 911  *
 912  * Set the VCPU's CPUID.
 913  */
 914 void vcpu_set_cpuid(struct kvm_vm *vm,
 915                 uint32_t vcpuid, struct kvm_cpuid2 *cpuid)
 916 {
 917         int rc;
 918
 919         rc = __vcpu_set_cpuid(vm, vcpuid, cpuid);
 920         TEST_ASSERT(rc == 0, "KVM_SET_CPUID2 failed, rc: %i errno: %i",
 921                     rc, errno);
 922
 923 }
 924
 925 /*
 926  * VCPU Get MSR
 927  *
 928  * Input Args:
 929  *   vm - Virtual Machine
 930  *   vcpuid - VCPU ID
 931  *   msr_index - Index of MSR
 932  *
 933  * Output Args: None
 934  *
 935  * Return: On success, value of the MSR. On failure a TEST_ASSERT is produced.
 936  *
 937  * Get value of MSR for VCPU.
 938  */
 939 uint64_t vcpu_get_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index)
 940 {
 941         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
 942         struct {
 943                 struct kvm_msrs header;
 944                 struct kvm_msr_entry entry;
 945         } buffer = {};
 946         int r;
 947
 948         TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid);
 949         buffer.header.nmsrs = 1;
 950         buffer.entry.index = msr_index;
 951         r = ioctl(vcpu->fd, KVM_GET_MSRS, &buffer.header);
 952         TEST_ASSERT(r == 1, "KVM_GET_MSRS IOCTL failed,\n"
 953                 "  rc: %i errno: %i", r, errno);
 954
 955         return buffer.entry.data;
 956 }
 957
 958 /*
 959  * _VCPU Set MSR
 960  *
 961  * Input Args:
 962  *   vm - Virtual Machine
 963  *   vcpuid - VCPU ID
 964  *   msr_index - Index of MSR
 965  *   msr_value - New value of MSR
 966  *
 967  * Output Args: None
 968  *
 969  * Return: The result of KVM_SET_MSRS.
 970  *
 971  * Sets the value of an MSR for the given VCPU.
 972  */
 973 int _vcpu_set_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index,
 974                   uint64_t msr_value)
 975 {
 976         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
 977         struct {
 978                 struct kvm_msrs header;
 979                 struct kvm_msr_entry entry;
 980         } buffer = {};
 981         int r;
 982
 983         TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid);
 984         memset(&buffer, 0, sizeof(buffer));
 985         buffer.header.nmsrs = 1;
 986         buffer.entry.index = msr_index;
 987         buffer.entry.data = msr_value;
 988         r = ioctl(vcpu->fd, KVM_SET_MSRS, &buffer.header);
 989         return r;
 990 }
 991
 992 /*
 993  * VCPU Set MSR
 994  *
 995  * Input Args:
 996  *   vm - Virtual Machine
 997  *   vcpuid - VCPU ID
 998  *   msr_index - Index of MSR
 999  *   msr_value - New value of MSR
1000  *
1001  * Output Args: None
1002  *
1003  * Return: On success, nothing. On failure a TEST_ASSERT is produced.
1004  *
1005  * Set value of MSR for VCPU.
1006  */
1007 void vcpu_set_msr(struct kvm_vm *vm, uint32_t vcpuid, uint64_t msr_index,
1008         uint64_t msr_value)
1009 {
1010         int r;
1011
1012         r = _vcpu_set_msr(vm, vcpuid, msr_index, msr_value);
1013         TEST_ASSERT(r == 1, "KVM_SET_MSRS IOCTL failed,\n"
1014                 "  rc: %i errno: %i", r, errno);
1015 }
1016
1017 void vcpu_args_set(struct kvm_vm *vm, uint32_t vcpuid, unsigned int num, ...)
1018 {
1019         va_list ap;
1020         struct kvm_regs regs;
1021
1022         TEST_ASSERT(num >= 1 && num <= 6, "Unsupported number of args,\n"
1023                     "  num: %u\n",
1024                     num);
1025
1026         va_start(ap, num);
1027         vcpu_regs_get(vm, vcpuid, &regs);
1028
1029         if (num >= 1)
1030                 regs.rdi = va_arg(ap, uint64_t);
1031
1032         if (num >= 2)
1033                 regs.rsi = va_arg(ap, uint64_t);
1034
1035         if (num >= 3)
1036                 regs.rdx = va_arg(ap, uint64_t);
1037
1038         if (num >= 4)
1039                 regs.rcx = va_arg(ap, uint64_t);
1040
1041         if (num >= 5)
1042                 regs.r8 = va_arg(ap, uint64_t);
1043
1044         if (num >= 6)
1045                 regs.r9 = va_arg(ap, uint64_t);
1046
1047         vcpu_regs_set(vm, vcpuid, &regs);
1048         va_end(ap);
1049 }
1050
1051 void vcpu_dump(FILE *stream, struct kvm_vm *vm, uint32_t vcpuid, uint8_t indent)
1052 {
1053         struct kvm_regs regs;
1054         struct kvm_sregs sregs;
1055
1056         fprintf(stream, "%*scpuid: %u\n", indent, "", vcpuid);
1057
1058         fprintf(stream, "%*sregs:\n", indent + 2, "");
1059         vcpu_regs_get(vm, vcpuid, &regs);
1060         regs_dump(stream, &regs, indent + 4);
1061
1062         fprintf(stream, "%*ssregs:\n", indent + 2, "");
1063         vcpu_sregs_get(vm, vcpuid, &sregs);
1064         sregs_dump(stream, &sregs, indent + 4);
1065 }
1066
1067 static int kvm_get_num_msrs_fd(int kvm_fd)
1068 {
1069         struct kvm_msr_list nmsrs;
1070         int r;
1071
1072         nmsrs.nmsrs = 0;
1073         r = ioctl(kvm_fd, KVM_GET_MSR_INDEX_LIST, &nmsrs);
1074         TEST_ASSERT(r == -1 && errno == E2BIG, "Unexpected result from KVM_GET_MSR_INDEX_LIST probe, r: %i",
1075                 r);
1076
1077         return nmsrs.nmsrs;
1078 }
1079
1080 static int kvm_get_num_msrs(struct kvm_vm *vm)
1081 {
1082         return kvm_get_num_msrs_fd(vm->kvm_fd);
1083 }
1084
1085 struct kvm_msr_list *kvm_get_msr_index_list(void)
1086 {
1087         struct kvm_msr_list *list;
1088         int nmsrs, r, kvm_fd;
1089
1090         kvm_fd = open_kvm_dev_path_or_exit();
1091
1092         nmsrs = kvm_get_num_msrs_fd(kvm_fd);
1093         list = malloc(sizeof(*list) + nmsrs * sizeof(list->indices[0]));
1094         list->nmsrs = nmsrs;
1095         r = ioctl(kvm_fd, KVM_GET_MSR_INDEX_LIST, list);
1096         close(kvm_fd);
1097
1098         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MSR_INDEX_LIST, r: %i",
1099                 r);
1100
1101         return list;
1102 }
1103
1104 static int vcpu_save_xsave_state(struct kvm_vm *vm, struct vcpu *vcpu,
1105                                  struct kvm_x86_state *state)
1106 {
1107         int size;
1108
1109         size = vm_check_cap(vm, KVM_CAP_XSAVE2);
1110         if (!size)
1111                 size = sizeof(struct kvm_xsave);
1112
1113         state->xsave = malloc(size);
1114         if (size == sizeof(struct kvm_xsave))
1115                 return ioctl(vcpu->fd, KVM_GET_XSAVE, state->xsave);
1116         else
1117                 return ioctl(vcpu->fd, KVM_GET_XSAVE2, state->xsave);
1118 }
1119
1120 struct kvm_x86_state *vcpu_save_state(struct kvm_vm *vm, uint32_t vcpuid)
1121 {
1122         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
1123         struct kvm_msr_list *list;
1124         struct kvm_x86_state *state;
1125         int nmsrs, r, i;
1126         static int nested_size = -1;
1127
1128         if (nested_size == -1) {
1129                 nested_size = kvm_check_cap(KVM_CAP_NESTED_STATE);
1130                 TEST_ASSERT(nested_size <= sizeof(state->nested_),
1131                             "Nested state size too big, %i > %zi",
1132                             nested_size, sizeof(state->nested_));
1133         }
1134
1135         /*
1136          * When KVM exits to userspace with KVM_EXIT_IO, KVM guarantees
1137          * guest state is consistent only after userspace re-enters the
1138          * kernel with KVM_RUN.  Complete IO prior to migrating state
1139          * to a new VM.
1140          */
1141         vcpu_run_complete_io(vm, vcpuid);
1142
1143         nmsrs = kvm_get_num_msrs(vm);
1144         list = malloc(sizeof(*list) + nmsrs * sizeof(list->indices[0]));
1145         list->nmsrs = nmsrs;
1146         r = ioctl(vm->kvm_fd, KVM_GET_MSR_INDEX_LIST, list);
1147         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MSR_INDEX_LIST, r: %i",
1148                     r);
1149
1150         state = malloc(sizeof(*state) + nmsrs * sizeof(state->msrs.entries[0]));
1151         r = ioctl(vcpu->fd, KVM_GET_VCPU_EVENTS, &state->events);
1152         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_VCPU_EVENTS, r: %i",
1153                     r);
1154
1155         r = ioctl(vcpu->fd, KVM_GET_MP_STATE, &state->mp_state);
1156         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MP_STATE, r: %i",
1157                     r);
1158
1159         r = ioctl(vcpu->fd, KVM_GET_REGS, &state->regs);
1160         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_REGS, r: %i",
1161                     r);
1162
1163         r = vcpu_save_xsave_state(vm, vcpu, state);
1164         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_XSAVE, r: %i",
1165                     r);
1166
1167         if (kvm_check_cap(KVM_CAP_XCRS)) {
1168                 r = ioctl(vcpu->fd, KVM_GET_XCRS, &state->xcrs);
1169                 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_XCRS, r: %i",
1170                             r);
1171         }
1172
1173         r = ioctl(vcpu->fd, KVM_GET_SREGS, &state->sregs);
1174         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_SREGS, r: %i",
1175                     r);
1176
1177         if (nested_size) {
1178                 state->nested.size = sizeof(state->nested_);
1179                 r = ioctl(vcpu->fd, KVM_GET_NESTED_STATE, &state->nested);
1180                 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_NESTED_STATE, r: %i",
1181                             r);
1182                 TEST_ASSERT(state->nested.size <= nested_size,
1183                             "Nested state size too big, %i (KVM_CHECK_CAP gave %i)",
1184                             state->nested.size, nested_size);
1185         } else
1186                 state->nested.size = 0;
1187
1188         state->msrs.nmsrs = nmsrs;
1189         for (i = 0; i < nmsrs; i++)
1190                 state->msrs.entries[i].index = list->indices[i];
1191         r = ioctl(vcpu->fd, KVM_GET_MSRS, &state->msrs);
1192         TEST_ASSERT(r == nmsrs, "Unexpected result from KVM_GET_MSRS, r: %i (failed MSR was 0x%x)",
1193                     r, r == nmsrs ? -1 : list->indices[r]);
1194
1195         r = ioctl(vcpu->fd, KVM_GET_DEBUGREGS, &state->debugregs);
1196         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_DEBUGREGS, r: %i",
1197                     r);
1198
1199         free(list);
1200         return state;
1201 }
1202
1203 void vcpu_load_state(struct kvm_vm *vm, uint32_t vcpuid, struct kvm_x86_state *state)
1204 {
1205         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
1206         int r;
1207
1208         r = ioctl(vcpu->fd, KVM_SET_SREGS, &state->sregs);
1209         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_SREGS, r: %i",
1210                     r);
1211
1212         r = ioctl(vcpu->fd, KVM_SET_MSRS, &state->msrs);
1213         TEST_ASSERT(r == state->msrs.nmsrs,
1214                 "Unexpected result from KVM_SET_MSRS, r: %i (failed at %x)",
1215                 r, r == state->msrs.nmsrs ? -1 : state->msrs.entries[r].index);
1216
1217         if (kvm_check_cap(KVM_CAP_XCRS)) {
1218                 r = ioctl(vcpu->fd, KVM_SET_XCRS, &state->xcrs);
1219                 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_XCRS, r: %i",
1220                             r);
1221         }
1222
1223         r = ioctl(vcpu->fd, KVM_SET_XSAVE, state->xsave);
1224         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_XSAVE, r: %i",
1225                     r);
1226
1227         r = ioctl(vcpu->fd, KVM_SET_VCPU_EVENTS, &state->events);
1228         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_VCPU_EVENTS, r: %i",
1229                     r);
1230
1231         r = ioctl(vcpu->fd, KVM_SET_MP_STATE, &state->mp_state);
1232         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_MP_STATE, r: %i",
1233                     r);
1234
1235         r = ioctl(vcpu->fd, KVM_SET_DEBUGREGS, &state->debugregs);
1236         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_DEBUGREGS, r: %i",
1237                     r);
1238
1239         r = ioctl(vcpu->fd, KVM_SET_REGS, &state->regs);
1240         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_REGS, r: %i",
1241                     r);
1242
1243         if (state->nested.size) {
1244                 r = ioctl(vcpu->fd, KVM_SET_NESTED_STATE, &state->nested);
1245                 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_NESTED_STATE, r: %i",
1246                             r);
1247         }
1248 }
1249
1250 void kvm_x86_state_cleanup(struct kvm_x86_state *state)
1251 {
1252         free(state->xsave);
1253         free(state);
1254 }
1255
1256 static bool cpu_vendor_string_is(const char *vendor)
1257 {
1258         const uint32_t *chunk = (const uint32_t *)vendor;
1259         int eax, ebx, ecx, edx;
1260         const int leaf = 0;
1261
1262         __asm__ __volatile__(
1263                 "cpuid"
1264                 : /* output */ "=a"(eax), "=b"(ebx),
1265                   "=c"(ecx), "=d"(edx)
1266                 : /* input */ "0"(leaf), "2"(0));
1267
1268         return (ebx == chunk[0] && edx == chunk[1] && ecx == chunk[2]);
1269 }
1270
1271 bool is_intel_cpu(void)
1272 {
1273         return cpu_vendor_string_is("GenuineIntel");
1274 }
1275
1276 /*
1277  * Exclude early K5 samples with a vendor string of "AMDisbetter!"
1278  */
1279 bool is_amd_cpu(void)
1280 {
1281         return cpu_vendor_string_is("AuthenticAMD");
1282 }
1283
1284 uint32_t kvm_get_cpuid_max_basic(void)
1285 {
1286         return kvm_get_supported_cpuid_entry(0)->eax;
1287 }
1288
1289 uint32_t kvm_get_cpuid_max_extended(void)
1290 {
1291         return kvm_get_supported_cpuid_entry(0x80000000)->eax;
1292 }
1293
1294 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits)
1295 {
1296         struct kvm_cpuid_entry2 *entry;
1297         bool pae;
1298
1299         /* SDM 4.1.4 */
1300         if (kvm_get_cpuid_max_extended() < 0x80000008) {
1301                 pae = kvm_get_supported_cpuid_entry(1)->edx & (1 << 6);
1302                 *pa_bits = pae ? 36 : 32;
1303                 *va_bits = 32;
1304         } else {
1305                 entry = kvm_get_supported_cpuid_entry(0x80000008);
1306                 *pa_bits = entry->eax & 0xff;
1307                 *va_bits = (entry->eax >> 8) & 0xff;
1308         }
1309 }
1310
1311 struct idt_entry {
1312         uint16_t offset0;
1313         uint16_t selector;
1314         uint16_t ist : 3;
1315         uint16_t : 5;
1316         uint16_t type : 4;
1317         uint16_t : 1;
1318         uint16_t dpl : 2;
1319         uint16_t p : 1;
1320         uint16_t offset1;
1321         uint32_t offset2; uint32_t reserved;
1322 };
1323
1324 static void set_idt_entry(struct kvm_vm *vm, int vector, unsigned long addr,
1325                           int dpl, unsigned short selector)
1326 {
1327         struct idt_entry *base =
1328                 (struct idt_entry *)addr_gva2hva(vm, vm->idt);
1329         struct idt_entry *e = &base[vector];
1330
1331         memset(e, 0, sizeof(*e));
1332         e->offset0 = addr;
1333         e->selector = selector;
1334         e->ist = 0;
1335         e->type = 14;
1336         e->dpl = dpl;
1337         e->p = 1;
1338         e->offset1 = addr >> 16;
1339         e->offset2 = addr >> 32;
1340 }
1341
1342 void kvm_exit_unexpected_vector(uint32_t value)
1343 {
1344         ucall(UCALL_UNHANDLED, 1, value);
1345 }
1346
1347 void route_exception(struct ex_regs *regs)
1348 {
1349         typedef void(*handler)(struct ex_regs *);
1350         handler *handlers = (handler *)exception_handlers;
1351
1352         if (handlers && handlers[regs->vector]) {
1353                 handlers[regs->vector](regs);
1354                 return;
1355         }
1356
1357         kvm_exit_unexpected_vector(regs->vector);
1358 }
1359
1360 void vm_init_descriptor_tables(struct kvm_vm *vm)
1361 {
1362         extern void *idt_handlers;
1363         int i;
1364
1365         vm->idt = vm_vaddr_alloc_page(vm);
1366         vm->handlers = vm_vaddr_alloc_page(vm);
1367         /* Handlers have the same address in both address spaces.*/
1368         for (i = 0; i < NUM_INTERRUPTS; i++)
1369                 set_idt_entry(vm, i, (unsigned long)(&idt_handlers)[i], 0,
1370                         DEFAULT_CODE_SELECTOR);
1371 }
1372
1373 void vcpu_init_descriptor_tables(struct kvm_vm *vm, uint32_t vcpuid)
1374 {
1375         struct kvm_sregs sregs;
1376
1377         vcpu_sregs_get(vm, vcpuid, &sregs);
1378         sregs.idt.base = vm->idt;
1379         sregs.idt.limit = NUM_INTERRUPTS * sizeof(struct idt_entry) - 1;
1380         sregs.gdt.base = vm->gdt;
1381         sregs.gdt.limit = getpagesize() - 1;
1382         kvm_seg_set_kernel_data_64bit(NULL, DEFAULT_DATA_SELECTOR, &sregs.gs);
1383         vcpu_sregs_set(vm, vcpuid, &sregs);
1384         *(vm_vaddr_t *)addr_gva2hva(vm, (vm_vaddr_t)(&exception_handlers)) = vm->handlers;
1385 }
1386
1387 void vm_install_exception_handler(struct kvm_vm *vm, int vector,
1388                                void (*handler)(struct ex_regs *))
1389 {
1390         vm_vaddr_t *handlers = (vm_vaddr_t *)addr_gva2hva(vm, vm->handlers);
1391
1392         handlers[vector] = (vm_vaddr_t)handler;
1393 }
1394
1395 void assert_on_unhandled_exception(struct kvm_vm *vm, uint32_t vcpuid)
1396 {
1397         struct ucall uc;
1398
1399         if (get_ucall(vm, vcpuid, &uc) == UCALL_UNHANDLED) {
1400                 uint64_t vector = uc.args[0];
1401
1402                 TEST_FAIL("Unexpected vectored event in guest (vector:0x%lx)",
1403                           vector);
1404         }
1405 }
1406
1407 struct kvm_cpuid_entry2 *get_cpuid(struct kvm_cpuid2 *cpuid, uint32_t function,
1408                                    uint32_t index)
1409 {
1410         int i;
1411
1412         for (i = 0; i < cpuid->nent; i++) {
1413                 struct kvm_cpuid_entry2 *cur = &cpuid->entries[i];
1414
1415                 if (cur->function == function && cur->index == index)
1416                         return cur;
1417         }
1418
1419         TEST_FAIL("CPUID function 0x%x index 0x%x not found ", function, index);
1420
1421         return NULL;
1422 }
1423
1424 bool set_cpuid(struct kvm_cpuid2 *cpuid,
1425                struct kvm_cpuid_entry2 *ent)
1426 {
1427         int i;
1428
1429         for (i = 0; i < cpuid->nent; i++) {
1430                 struct kvm_cpuid_entry2 *cur = &cpuid->entries[i];
1431
1432                 if (cur->function != ent->function || cur->index != ent->index)
1433                         continue;
1434
1435                 memcpy(cur, ent, sizeof(struct kvm_cpuid_entry2));
1436                 return true;
1437         }
1438
1439         return false;
1440 }
1441
1442 uint64_t kvm_hypercall(uint64_t nr, uint64_t a0, uint64_t a1, uint64_t a2,
1443                        uint64_t a3)
1444 {
1445         uint64_t r;
1446
1447         asm volatile("vmcall"
1448                      : "=a"(r)
1449                      : "b"(a0), "c"(a1), "d"(a2), "S"(a3));
1450         return r;
1451 }
1452
1453 struct kvm_cpuid2 *kvm_get_supported_hv_cpuid(void)
1454 {
1455         static struct kvm_cpuid2 *cpuid;
1456         int ret;
1457         int kvm_fd;
1458
1459         if (cpuid)
1460                 return cpuid;
1461
1462         cpuid = allocate_kvm_cpuid2();
1463         kvm_fd = open_kvm_dev_path_or_exit();
1464
1465         ret = ioctl(kvm_fd, KVM_GET_SUPPORTED_HV_CPUID, cpuid);
1466         TEST_ASSERT(ret == 0, "KVM_GET_SUPPORTED_HV_CPUID failed %d %d\n",
1467                     ret, errno);
1468
1469         close(kvm_fd);
1470         return cpuid;
1471 }
1472
1473 void vcpu_set_hv_cpuid(struct kvm_vm *vm, uint32_t vcpuid)
1474 {
1475         static struct kvm_cpuid2 *cpuid_full;
1476         struct kvm_cpuid2 *cpuid_sys, *cpuid_hv;
1477         int i, nent = 0;
1478
1479         if (!cpuid_full) {
1480                 cpuid_sys = kvm_get_supported_cpuid();
1481                 cpuid_hv = kvm_get_supported_hv_cpuid();
1482
1483                 cpuid_full = malloc(sizeof(*cpuid_full) +
1484                                     (cpuid_sys->nent + cpuid_hv->nent) *
1485                                     sizeof(struct kvm_cpuid_entry2));
1486                 if (!cpuid_full) {
1487                         perror("malloc");
1488                         abort();
1489                 }
1490
1491                 /* Need to skip KVM CPUID leaves 0x400000xx */
1492                 for (i = 0; i < cpuid_sys->nent; i++) {
1493                         if (cpuid_sys->entries[i].function >= 0x40000000 &&
1494                             cpuid_sys->entries[i].function < 0x40000100)
1495                                 continue;
1496                         cpuid_full->entries[nent] = cpuid_sys->entries[i];
1497                         nent++;
1498                 }
1499
1500                 memcpy(&cpuid_full->entries[nent], cpuid_hv->entries,
1501                        cpuid_hv->nent * sizeof(struct kvm_cpuid_entry2));
1502                 cpuid_full->nent = nent + cpuid_hv->nent;
1503         }
1504
1505         vcpu_set_cpuid(vm, vcpuid, cpuid_full);
1506 }
1507
1508 struct kvm_cpuid2 *vcpu_get_supported_hv_cpuid(struct kvm_vm *vm, uint32_t vcpuid)
1509 {
1510         static struct kvm_cpuid2 *cpuid;
1511
1512         cpuid = allocate_kvm_cpuid2();
1513
1514         vcpu_ioctl(vm, vcpuid, KVM_GET_SUPPORTED_HV_CPUID, cpuid);
1515
1516         return cpuid;
1517 }
1518
1519 unsigned long vm_compute_max_gfn(struct kvm_vm *vm)
1520 {
1521         const unsigned long num_ht_pages = 12 << (30 - vm->page_shift); /* 12 GiB */
1522         unsigned long ht_gfn, max_gfn, max_pfn;
1523         uint32_t eax, ebx, ecx, edx, max_ext_leaf;
1524
1525         max_gfn = (1ULL << (vm->pa_bits - vm->page_shift)) - 1;
1526
1527         /* Avoid reserved HyperTransport region on AMD processors.  */
1528         if (!is_amd_cpu())
1529                 return max_gfn;
1530
1531         /* On parts with <40 physical address bits, the area is fully hidden */
1532         if (vm->pa_bits < 40)
1533                 return max_gfn;
1534
1535         /* Before family 17h, the HyperTransport area is just below 1T.  */
1536         ht_gfn = (1 << 28) - num_ht_pages;
1537         eax = 1;
1538         ecx = 0;
1539         cpuid(&eax, &ebx, &ecx, &edx);
1540         if (x86_family(eax) < 0x17)
1541                 goto done;
1542
1543         /*
1544          * Otherwise it's at the top of the physical address space, possibly
1545          * reduced due to SME by bits 11:6 of CPUID[0x8000001f].EBX.  Use
1546          * the old conservative value if MAXPHYADDR is not enumerated.
1547          */
1548         eax = 0x80000000;
1549         cpuid(&eax, &ebx, &ecx, &edx);
1550         max_ext_leaf = eax;
1551         if (max_ext_leaf < 0x80000008)
1552                 goto done;
1553
1554         eax = 0x80000008;
1555         cpuid(&eax, &ebx, &ecx, &edx);
1556         max_pfn = (1ULL << ((eax & 0xff) - vm->page_shift)) - 1;
1557         if (max_ext_leaf >= 0x8000001f) {
1558                 eax = 0x8000001f;
1559                 cpuid(&eax, &ebx, &ecx, &edx);
1560                 max_pfn >>= (ebx >> 6) & 0x3f;
1561         }
1562
1563         ht_gfn = max_pfn - num_ht_pages;
1564 done:
1565         return min(max_gfn, ht_gfn - 1);
1566 }