tools/testing/selftests/kvm/lib/x86.c

   1 /*
   2  * tools/testing/selftests/kvm/lib/x86.c
   3  *
   4  * Copyright (C) 2018, Google LLC.
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2.
   7  */
   8
   9 #define _GNU_SOURCE /* for program_invocation_name */
  10
  11 #include "test_util.h"
  12 #include "kvm_util.h"
  13 #include "kvm_util_internal.h"
  14 #include "x86.h"
  15
  16 /* Minimum physical address used for virtual translation tables. */
  17 #define KVM_GUEST_PAGE_TABLE_MIN_PADDR 0x180000
  18
  19 /* Virtual translation table structure declarations */
  20 struct pageMapL4Entry {
  21         uint64_t present:1;
  22         uint64_t writable:1;
  23         uint64_t user:1;
  24         uint64_t write_through:1;
  25         uint64_t cache_disable:1;
  26         uint64_t accessed:1;
  27         uint64_t ignored_06:1;
  28         uint64_t page_size:1;
  29         uint64_t ignored_11_08:4;
  30         uint64_t address:40;
  31         uint64_t ignored_62_52:11;
  32         uint64_t execute_disable:1;
  33 };
  34
  35 struct pageDirectoryPointerEntry {
  36         uint64_t present:1;
  37         uint64_t writable:1;
  38         uint64_t user:1;
  39         uint64_t write_through:1;
  40         uint64_t cache_disable:1;
  41         uint64_t accessed:1;
  42         uint64_t ignored_06:1;
  43         uint64_t page_size:1;
  44         uint64_t ignored_11_08:4;
  45         uint64_t address:40;
  46         uint64_t ignored_62_52:11;
  47         uint64_t execute_disable:1;
  48 };
  49
  50 struct pageDirectoryEntry {
  51         uint64_t present:1;
  52         uint64_t writable:1;
  53         uint64_t user:1;
  54         uint64_t write_through:1;
  55         uint64_t cache_disable:1;
  56         uint64_t accessed:1;
  57         uint64_t ignored_06:1;
  58         uint64_t page_size:1;
  59         uint64_t ignored_11_08:4;
  60         uint64_t address:40;
  61         uint64_t ignored_62_52:11;
  62         uint64_t execute_disable:1;
  63 };
  64
  65 struct pageTableEntry {
  66         uint64_t present:1;
  67         uint64_t writable:1;
  68         uint64_t user:1;
  69         uint64_t write_through:1;
  70         uint64_t cache_disable:1;
  71         uint64_t accessed:1;
  72         uint64_t dirty:1;
  73         uint64_t reserved_07:1;
  74         uint64_t global:1;
  75         uint64_t ignored_11_09:3;
  76         uint64_t address:40;
  77         uint64_t ignored_62_52:11;
  78         uint64_t execute_disable:1;
  79 };
  80
  81 /* Register Dump
  82  *
  83  * Input Args:
  84  *   indent - Left margin indent amount
  85  *   regs - register
  86  *
  87  * Output Args:
  88  *   stream - Output FILE stream
  89  *
  90  * Return: None
  91  *
  92  * Dumps the state of the registers given by regs, to the FILE stream
  93  * given by steam.
  94  */
  95 void regs_dump(FILE *stream, struct kvm_regs *regs,
  96                uint8_t indent)
  97 {
  98         fprintf(stream, "%*srax: 0x%.16llx rbx: 0x%.16llx "
  99                 "rcx: 0x%.16llx rdx: 0x%.16llx\n",
 100                 indent, "",
 101                 regs->rax, regs->rbx, regs->rcx, regs->rdx);
 102         fprintf(stream, "%*srsi: 0x%.16llx rdi: 0x%.16llx "
 103                 "rsp: 0x%.16llx rbp: 0x%.16llx\n",
 104                 indent, "",
 105                 regs->rsi, regs->rdi, regs->rsp, regs->rbp);
 106         fprintf(stream, "%*sr8:  0x%.16llx r9:  0x%.16llx "
 107                 "r10: 0x%.16llx r11: 0x%.16llx\n",
 108                 indent, "",
 109                 regs->r8, regs->r9, regs->r10, regs->r11);
 110         fprintf(stream, "%*sr12: 0x%.16llx r13: 0x%.16llx "
 111                 "r14: 0x%.16llx r15: 0x%.16llx\n",
 112                 indent, "",
 113                 regs->r12, regs->r13, regs->r14, regs->r15);
 114         fprintf(stream, "%*srip: 0x%.16llx rfl: 0x%.16llx\n",
 115                 indent, "",
 116                 regs->rip, regs->rflags);
 117 }
 118
 119 /* Segment Dump
 120  *
 121  * Input Args:
 122  *   indent - Left margin indent amount
 123  *   segment - KVM segment
 124  *
 125  * Output Args:
 126  *   stream - Output FILE stream
 127  *
 128  * Return: None
 129  *
 130  * Dumps the state of the KVM segment given by segment, to the FILE stream
 131  * given by steam.
 132  */
 133 static void segment_dump(FILE *stream, struct kvm_segment *segment,
 134                          uint8_t indent)
 135 {
 136         fprintf(stream, "%*sbase: 0x%.16llx limit: 0x%.8x "
 137                 "selector: 0x%.4x type: 0x%.2x\n",
 138                 indent, "", segment->base, segment->limit,
 139                 segment->selector, segment->type);
 140         fprintf(stream, "%*spresent: 0x%.2x dpl: 0x%.2x "
 141                 "db: 0x%.2x s: 0x%.2x l: 0x%.2x\n",
 142                 indent, "", segment->present, segment->dpl,
 143                 segment->db, segment->s, segment->l);
 144         fprintf(stream, "%*sg: 0x%.2x avl: 0x%.2x "
 145                 "unusable: 0x%.2x padding: 0x%.2x\n",
 146                 indent, "", segment->g, segment->avl,
 147                 segment->unusable, segment->padding);
 148 }
 149
 150 /* dtable Dump
 151  *
 152  * Input Args:
 153  *   indent - Left margin indent amount
 154  *   dtable - KVM dtable
 155  *
 156  * Output Args:
 157  *   stream - Output FILE stream
 158  *
 159  * Return: None
 160  *
 161  * Dumps the state of the KVM dtable given by dtable, to the FILE stream
 162  * given by steam.
 163  */
 164 static void dtable_dump(FILE *stream, struct kvm_dtable *dtable,
 165                         uint8_t indent)
 166 {
 167         fprintf(stream, "%*sbase: 0x%.16llx limit: 0x%.4x "
 168                 "padding: 0x%.4x 0x%.4x 0x%.4x\n",
 169                 indent, "", dtable->base, dtable->limit,
 170                 dtable->padding[0], dtable->padding[1], dtable->padding[2]);
 171 }
 172
 173 /* System Register Dump
 174  *
 175  * Input Args:
 176  *   indent - Left margin indent amount
 177  *   sregs - System registers
 178  *
 179  * Output Args:
 180  *   stream - Output FILE stream
 181  *
 182  * Return: None
 183  *
 184  * Dumps the state of the system registers given by sregs, to the FILE stream
 185  * given by steam.
 186  */
 187 void sregs_dump(FILE *stream, struct kvm_sregs *sregs,
 188                 uint8_t indent)
 189 {
 190         unsigned int i;
 191
 192         fprintf(stream, "%*scs:\n", indent, "");
 193         segment_dump(stream, &sregs->cs, indent + 2);
 194         fprintf(stream, "%*sds:\n", indent, "");
 195         segment_dump(stream, &sregs->ds, indent + 2);
 196         fprintf(stream, "%*ses:\n", indent, "");
 197         segment_dump(stream, &sregs->es, indent + 2);
 198         fprintf(stream, "%*sfs:\n", indent, "");
 199         segment_dump(stream, &sregs->fs, indent + 2);
 200         fprintf(stream, "%*sgs:\n", indent, "");
 201         segment_dump(stream, &sregs->gs, indent + 2);
 202         fprintf(stream, "%*sss:\n", indent, "");
 203         segment_dump(stream, &sregs->ss, indent + 2);
 204         fprintf(stream, "%*str:\n", indent, "");
 205         segment_dump(stream, &sregs->tr, indent + 2);
 206         fprintf(stream, "%*sldt:\n", indent, "");
 207         segment_dump(stream, &sregs->ldt, indent + 2);
 208
 209         fprintf(stream, "%*sgdt:\n", indent, "");
 210         dtable_dump(stream, &sregs->gdt, indent + 2);
 211         fprintf(stream, "%*sidt:\n", indent, "");
 212         dtable_dump(stream, &sregs->idt, indent + 2);
 213
 214         fprintf(stream, "%*scr0: 0x%.16llx cr2: 0x%.16llx "
 215                 "cr3: 0x%.16llx cr4: 0x%.16llx\n",
 216                 indent, "",
 217                 sregs->cr0, sregs->cr2, sregs->cr3, sregs->cr4);
 218         fprintf(stream, "%*scr8: 0x%.16llx efer: 0x%.16llx "
 219                 "apic_base: 0x%.16llx\n",
 220                 indent, "",
 221                 sregs->cr8, sregs->efer, sregs->apic_base);
 222
 223         fprintf(stream, "%*sinterrupt_bitmap:\n", indent, "");
 224         for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++) {
 225                 fprintf(stream, "%*s%.16llx\n", indent + 2, "",
 226                         sregs->interrupt_bitmap[i]);
 227         }
 228 }
 229
 230 void virt_pgd_alloc(struct kvm_vm *vm, uint32_t pgd_memslot)
 231 {
 232         int rc;
 233
 234         TEST_ASSERT(vm->mode == VM_MODE_FLAT48PG, "Attempt to use "
 235                 "unknown or unsupported guest mode, mode: 0x%x", vm->mode);
 236
 237         /* If needed, create page map l4 table. */
 238         if (!vm->pgd_created) {
 239                 vm_paddr_t paddr = vm_phy_page_alloc(vm,
 240                         KVM_GUEST_PAGE_TABLE_MIN_PADDR, pgd_memslot);
 241                 vm->pgd = paddr;
 242                 vm->pgd_created = true;
 243         }
 244 }
 245
 246 /* VM Virtual Page Map
 247  *
 248  * Input Args:
 249  *   vm - Virtual Machine
 250  *   vaddr - VM Virtual Address
 251  *   paddr - VM Physical Address
 252  *   pgd_memslot - Memory region slot for new virtual translation tables
 253  *
 254  * Output Args: None
 255  *
 256  * Return: None
 257  *
 258  * Within the VM given by vm, creates a virtual translation for the page
 259  * starting at vaddr to the page starting at paddr.
 260  */
 261 void virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
 262         uint32_t pgd_memslot)
 263 {
 264         uint16_t index[4];
 265         struct pageMapL4Entry *pml4e;
 266
 267         TEST_ASSERT(vm->mode == VM_MODE_FLAT48PG, "Attempt to use "
 268                 "unknown or unsupported guest mode, mode: 0x%x", vm->mode);
 269
 270         TEST_ASSERT((vaddr % vm->page_size) == 0,
 271                 "Virtual address not on page boundary,\n"
 272                 "  vaddr: 0x%lx vm->page_size: 0x%x",
 273                 vaddr, vm->page_size);
 274         TEST_ASSERT(sparsebit_is_set(vm->vpages_valid,
 275                 (vaddr >> vm->page_shift)),
 276                 "Invalid virtual address, vaddr: 0x%lx",
 277                 vaddr);
 278         TEST_ASSERT((paddr % vm->page_size) == 0,
 279                 "Physical address not on page boundary,\n"
 280                 "  paddr: 0x%lx vm->page_size: 0x%x",
 281                 paddr, vm->page_size);
 282         TEST_ASSERT((paddr >> vm->page_shift) <= vm->max_gfn,
 283                 "Physical address beyond beyond maximum supported,\n"
 284                 "  paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x",
 285                 paddr, vm->max_gfn, vm->page_size);
 286
 287         index[0] = (vaddr >> 12) & 0x1ffu;
 288         index[1] = (vaddr >> 21) & 0x1ffu;
 289         index[2] = (vaddr >> 30) & 0x1ffu;
 290         index[3] = (vaddr >> 39) & 0x1ffu;
 291
 292         /* Allocate page directory pointer table if not present. */
 293         pml4e = addr_gpa2hva(vm, vm->pgd);
 294         if (!pml4e[index[3]].present) {
 295                 pml4e[index[3]].address = vm_phy_page_alloc(vm,
 296                         KVM_GUEST_PAGE_TABLE_MIN_PADDR, pgd_memslot)
 297                         >> vm->page_shift;
 298                 pml4e[index[3]].writable = true;
 299                 pml4e[index[3]].present = true;
 300         }
 301
 302         /* Allocate page directory table if not present. */
 303         struct pageDirectoryPointerEntry *pdpe;
 304         pdpe = addr_gpa2hva(vm, pml4e[index[3]].address * vm->page_size);
 305         if (!pdpe[index[2]].present) {
 306                 pdpe[index[2]].address = vm_phy_page_alloc(vm,
 307                         KVM_GUEST_PAGE_TABLE_MIN_PADDR, pgd_memslot)
 308                         >> vm->page_shift;
 309                 pdpe[index[2]].writable = true;
 310                 pdpe[index[2]].present = true;
 311         }
 312
 313         /* Allocate page table if not present. */
 314         struct pageDirectoryEntry *pde;
 315         pde = addr_gpa2hva(vm, pdpe[index[2]].address * vm->page_size);
 316         if (!pde[index[1]].present) {
 317                 pde[index[1]].address = vm_phy_page_alloc(vm,
 318                         KVM_GUEST_PAGE_TABLE_MIN_PADDR, pgd_memslot)
 319                         >> vm->page_shift;
 320                 pde[index[1]].writable = true;
 321                 pde[index[1]].present = true;
 322         }
 323
 324         /* Fill in page table entry. */
 325         struct pageTableEntry *pte;
 326         pte = addr_gpa2hva(vm, pde[index[1]].address * vm->page_size);
 327         pte[index[0]].address = paddr >> vm->page_shift;
 328         pte[index[0]].writable = true;
 329         pte[index[0]].present = 1;
 330 }
 331
 332 /* Virtual Translation Tables Dump
 333  *
 334  * Input Args:
 335  *   vm - Virtual Machine
 336  *   indent - Left margin indent amount
 337  *
 338  * Output Args:
 339  *   stream - Output FILE stream
 340  *
 341  * Return: None
 342  *
 343  * Dumps to the FILE stream given by stream, the contents of all the
 344  * virtual translation tables for the VM given by vm.
 345  */
 346 void virt_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent)
 347 {
 348         struct pageMapL4Entry *pml4e, *pml4e_start;
 349         struct pageDirectoryPointerEntry *pdpe, *pdpe_start;
 350         struct pageDirectoryEntry *pde, *pde_start;
 351         struct pageTableEntry *pte, *pte_start;
 352
 353         if (!vm->pgd_created)
 354                 return;
 355
 356         fprintf(stream, "%*s                                          "
 357                 "                no\n", indent, "");
 358         fprintf(stream, "%*s      index hvaddr         gpaddr         "
 359                 "addr         w exec dirty\n",
 360                 indent, "");
 361         pml4e_start = (struct pageMapL4Entry *) addr_gpa2hva(vm,
 362                 vm->pgd);
 363         for (uint16_t n1 = 0; n1 <= 0x1ffu; n1++) {
 364                 pml4e = &pml4e_start[n1];
 365                 if (!pml4e->present)
 366                         continue;
 367                 fprintf(stream, "%*spml4e 0x%-3zx %p 0x%-12lx 0x%-10lx %u "
 368                         " %u\n",
 369                         indent, "",
 370                         pml4e - pml4e_start, pml4e,
 371                         addr_hva2gpa(vm, pml4e), (uint64_t) pml4e->address,
 372                         pml4e->writable, pml4e->execute_disable);
 373
 374                 pdpe_start = addr_gpa2hva(vm, pml4e->address
 375                         * vm->page_size);
 376                 for (uint16_t n2 = 0; n2 <= 0x1ffu; n2++) {
 377                         pdpe = &pdpe_start[n2];
 378                         if (!pdpe->present)
 379                                 continue;
 380                         fprintf(stream, "%*spdpe  0x%-3zx %p 0x%-12lx 0x%-10lx "
 381                                 "%u  %u\n",
 382                                 indent, "",
 383                                 pdpe - pdpe_start, pdpe,
 384                                 addr_hva2gpa(vm, pdpe),
 385                                 (uint64_t) pdpe->address, pdpe->writable,
 386                                 pdpe->execute_disable);
 387
 388                         pde_start = addr_gpa2hva(vm,
 389                                 pdpe->address * vm->page_size);
 390                         for (uint16_t n3 = 0; n3 <= 0x1ffu; n3++) {
 391                                 pde = &pde_start[n3];
 392                                 if (!pde->present)
 393                                         continue;
 394                                 fprintf(stream, "%*spde   0x%-3zx %p "
 395                                         "0x%-12lx 0x%-10lx %u  %u\n",
 396                                         indent, "", pde - pde_start, pde,
 397                                         addr_hva2gpa(vm, pde),
 398                                         (uint64_t) pde->address, pde->writable,
 399                                         pde->execute_disable);
 400
 401                                 pte_start = addr_gpa2hva(vm,
 402                                         pde->address * vm->page_size);
 403                                 for (uint16_t n4 = 0; n4 <= 0x1ffu; n4++) {
 404                                         pte = &pte_start[n4];
 405                                         if (!pte->present)
 406                                                 continue;
 407                                         fprintf(stream, "%*spte   0x%-3zx %p "
 408                                                 "0x%-12lx 0x%-10lx %u  %u "
 409                                                 "    %u    0x%-10lx\n",
 410                                                 indent, "",
 411                                                 pte - pte_start, pte,
 412                                                 addr_hva2gpa(vm, pte),
 413                                                 (uint64_t) pte->address,
 414                                                 pte->writable,
 415                                                 pte->execute_disable,
 416                                                 pte->dirty,
 417                                                 ((uint64_t) n1 << 27)
 418                                                         | ((uint64_t) n2 << 18)
 419                                                         | ((uint64_t) n3 << 9)
 420                                                         | ((uint64_t) n4));
 421                                 }
 422                         }
 423                 }
 424         }
 425 }
 426
 427 /* Set Unusable Segment
 428  *
 429  * Input Args: None
 430  *
 431  * Output Args:
 432  *   segp - Pointer to segment register
 433  *
 434  * Return: None
 435  *
 436  * Sets the segment register pointed to by segp to an unusable state.
 437  */
 438 static void kvm_seg_set_unusable(struct kvm_segment *segp)
 439 {
 440         memset(segp, 0, sizeof(*segp));
 441         segp->unusable = true;
 442 }
 443
 444 static void kvm_seg_fill_gdt_64bit(struct kvm_vm *vm, struct kvm_segment *segp)
 445 {
 446         void *gdt = addr_gva2hva(vm, vm->gdt);
 447         struct desc64 *desc = gdt + (segp->selector >> 3) * 8;
 448
 449         desc->limit0 = segp->limit & 0xFFFF;
 450         desc->base0 = segp->base & 0xFFFF;
 451         desc->base1 = segp->base >> 16;
 452         desc->type = segp->type;
 453         desc->s = segp->s;
 454         desc->dpl = segp->dpl;
 455         desc->p = segp->present;
 456         desc->limit1 = segp->limit >> 16;
 457         desc->avl = segp->avl;
 458         desc->l = segp->l;
 459         desc->db = segp->db;
 460         desc->g = segp->g;
 461         desc->base2 = segp->base >> 24;
 462         if (!segp->s)
 463                 desc->base3 = segp->base >> 32;
 464 }
 465
 466
 467 /* Set Long Mode Flat Kernel Code Segment
 468  *
 469  * Input Args:
 470  *   vm - VM whose GDT is being filled, or NULL to only write segp
 471  *   selector - selector value
 472  *
 473  * Output Args:
 474  *   segp - Pointer to KVM segment
 475  *
 476  * Return: None
 477  *
 478  * Sets up the KVM segment pointed to by segp, to be a code segment
 479  * with the selector value given by selector.
 480  */
 481 static void kvm_seg_set_kernel_code_64bit(struct kvm_vm *vm, uint16_t selector,
 482         struct kvm_segment *segp)
 483 {
 484         memset(segp, 0, sizeof(*segp));
 485         segp->selector = selector;
 486         segp->limit = 0xFFFFFFFFu;
 487         segp->s = 0x1; /* kTypeCodeData */
 488         segp->type = 0x08 | 0x01 | 0x02; /* kFlagCode | kFlagCodeAccessed
 489                                           * | kFlagCodeReadable
 490                                           */
 491         segp->g = true;
 492         segp->l = true;
 493         segp->present = 1;
 494         if (vm)
 495                 kvm_seg_fill_gdt_64bit(vm, segp);
 496 }
 497
 498 /* Set Long Mode Flat Kernel Data Segment
 499  *
 500  * Input Args:
 501  *   vm - VM whose GDT is being filled, or NULL to only write segp
 502  *   selector - selector value
 503  *
 504  * Output Args:
 505  *   segp - Pointer to KVM segment
 506  *
 507  * Return: None
 508  *
 509  * Sets up the KVM segment pointed to by segp, to be a data segment
 510  * with the selector value given by selector.
 511  */
 512 static void kvm_seg_set_kernel_data_64bit(struct kvm_vm *vm, uint16_t selector,
 513         struct kvm_segment *segp)
 514 {
 515         memset(segp, 0, sizeof(*segp));
 516         segp->selector = selector;
 517         segp->limit = 0xFFFFFFFFu;
 518         segp->s = 0x1; /* kTypeCodeData */
 519         segp->type = 0x00 | 0x01 | 0x02; /* kFlagData | kFlagDataAccessed
 520                                           * | kFlagDataWritable
 521                                           */
 522         segp->g = true;
 523         segp->present = true;
 524         if (vm)
 525                 kvm_seg_fill_gdt_64bit(vm, segp);
 526 }
 527
 528 /* Address Guest Virtual to Guest Physical
 529  *
 530  * Input Args:
 531  *   vm - Virtual Machine
 532  *   gpa - VM virtual address
 533  *
 534  * Output Args: None
 535  *
 536  * Return:
 537  *   Equivalent VM physical address
 538  *
 539  * Translates the VM virtual address given by gva to a VM physical
 540  * address and then locates the memory region containing the VM
 541  * physical address, within the VM given by vm.  When found, the host
 542  * virtual address providing the memory to the vm physical address is returned.
 543  * A TEST_ASSERT failure occurs if no region containing translated
 544  * VM virtual address exists.
 545  */
 546 vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
 547 {
 548         uint16_t index[4];
 549         struct pageMapL4Entry *pml4e;
 550         struct pageDirectoryPointerEntry *pdpe;
 551         struct pageDirectoryEntry *pde;
 552         struct pageTableEntry *pte;
 553         void *hva;
 554
 555         TEST_ASSERT(vm->mode == VM_MODE_FLAT48PG, "Attempt to use "
 556                 "unknown or unsupported guest mode, mode: 0x%x", vm->mode);
 557
 558         index[0] = (gva >> 12) & 0x1ffu;
 559         index[1] = (gva >> 21) & 0x1ffu;
 560         index[2] = (gva >> 30) & 0x1ffu;
 561         index[3] = (gva >> 39) & 0x1ffu;
 562
 563         if (!vm->pgd_created)
 564                 goto unmapped_gva;
 565         pml4e = addr_gpa2hva(vm, vm->pgd);
 566         if (!pml4e[index[3]].present)
 567                 goto unmapped_gva;
 568
 569         pdpe = addr_gpa2hva(vm, pml4e[index[3]].address * vm->page_size);
 570         if (!pdpe[index[2]].present)
 571                 goto unmapped_gva;
 572
 573         pde = addr_gpa2hva(vm, pdpe[index[2]].address * vm->page_size);
 574         if (!pde[index[1]].present)
 575                 goto unmapped_gva;
 576
 577         pte = addr_gpa2hva(vm, pde[index[1]].address * vm->page_size);
 578         if (!pte[index[0]].present)
 579                 goto unmapped_gva;
 580
 581         return (pte[index[0]].address * vm->page_size) + (gva & 0xfffu);
 582
 583 unmapped_gva:
 584         TEST_ASSERT(false, "No mapping for vm virtual address, "
 585                     "gva: 0x%lx", gva);
 586 }
 587
 588 static void kvm_setup_gdt(struct kvm_vm *vm, struct kvm_dtable *dt, int gdt_memslot,
 589                           int pgd_memslot)
 590 {
 591         if (!vm->gdt)
 592                 vm->gdt = vm_vaddr_alloc(vm, getpagesize(),
 593                         KVM_UTIL_MIN_VADDR, gdt_memslot, pgd_memslot);
 594
 595         dt->base = vm->gdt;
 596         dt->limit = getpagesize();
 597 }
 598
 599 static void kvm_setup_tss_64bit(struct kvm_vm *vm, struct kvm_segment *segp,
 600                                 int selector, int gdt_memslot,
 601                                 int pgd_memslot)
 602 {
 603         if (!vm->tss)
 604                 vm->tss = vm_vaddr_alloc(vm, getpagesize(),
 605                         KVM_UTIL_MIN_VADDR, gdt_memslot, pgd_memslot);
 606
 607         memset(segp, 0, sizeof(*segp));
 608         segp->base = vm->tss;
 609         segp->limit = 0x67;
 610         segp->selector = selector;
 611         segp->type = 0xb;
 612         segp->present = 1;
 613         kvm_seg_fill_gdt_64bit(vm, segp);
 614 }
 615
 616 void vcpu_setup(struct kvm_vm *vm, int vcpuid, int pgd_memslot, int gdt_memslot)
 617 {
 618         struct kvm_sregs sregs;
 619
 620         /* Set mode specific system register values. */
 621         vcpu_sregs_get(vm, vcpuid, &sregs);
 622
 623         sregs.idt.limit = 0;
 624
 625         kvm_setup_gdt(vm, &sregs.gdt, gdt_memslot, pgd_memslot);
 626
 627         switch (vm->mode) {
 628         case VM_MODE_FLAT48PG:
 629                 sregs.cr0 = X86_CR0_PE | X86_CR0_NE | X86_CR0_PG;
 630                 sregs.cr4 |= X86_CR4_PAE;
 631                 sregs.efer |= (EFER_LME | EFER_LMA | EFER_NX);
 632
 633                 kvm_seg_set_unusable(&sregs.ldt);
 634                 kvm_seg_set_kernel_code_64bit(vm, 0x8, &sregs.cs);
 635                 kvm_seg_set_kernel_data_64bit(vm, 0x10, &sregs.ds);
 636                 kvm_seg_set_kernel_data_64bit(vm, 0x10, &sregs.es);
 637                 kvm_setup_tss_64bit(vm, &sregs.tr, 0x18, gdt_memslot, pgd_memslot);
 638                 break;
 639
 640         default:
 641                 TEST_ASSERT(false, "Unknown guest mode, mode: 0x%x", vm->mode);
 642         }
 643
 644         sregs.cr3 = vm->pgd;
 645         vcpu_sregs_set(vm, vcpuid, &sregs);
 646 }
 647 /* Adds a vCPU with reasonable defaults (i.e., a stack)
 648  *
 649  * Input Args:
 650  *   vcpuid - The id of the VCPU to add to the VM.
 651  *   guest_code - The vCPU's entry point
 652  */
 653 void vm_vcpu_add_default(struct kvm_vm *vm, uint32_t vcpuid, void *guest_code)
 654 {
 655         struct kvm_mp_state mp_state;
 656         struct kvm_regs regs;
 657         vm_vaddr_t stack_vaddr;
 658         stack_vaddr = vm_vaddr_alloc(vm, DEFAULT_STACK_PGS * getpagesize(),
 659                                      DEFAULT_GUEST_STACK_VADDR_MIN, 0, 0);
 660
 661         /* Create VCPU */
 662         vm_vcpu_add(vm, vcpuid, 0, 0);
 663
 664         /* Setup guest general purpose registers */
 665         vcpu_regs_get(vm, vcpuid, &regs);
 666         regs.rflags = regs.rflags | 0x2;
 667         regs.rsp = stack_vaddr + (DEFAULT_STACK_PGS * getpagesize());
 668         regs.rip = (unsigned long) guest_code;
 669         vcpu_regs_set(vm, vcpuid, &regs);
 670
 671         /* Setup the MP state */
 672         mp_state.mp_state = 0;
 673         vcpu_set_mp_state(vm, vcpuid, &mp_state);
 674 }
 675
 676 /* VM VCPU CPUID Set
 677  *
 678  * Input Args:
 679  *   vm - Virtual Machine
 680  *   vcpuid - VCPU id
 681  *   cpuid - The CPUID values to set.
 682  *
 683  * Output Args: None
 684  *
 685  * Return: void
 686  *
 687  * Set the VCPU's CPUID.
 688  */
 689 void vcpu_set_cpuid(struct kvm_vm *vm,
 690                 uint32_t vcpuid, struct kvm_cpuid2 *cpuid)
 691 {
 692         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
 693         int rc;
 694
 695         TEST_ASSERT(vcpu != NULL, "vcpu not found, vcpuid: %u", vcpuid);
 696
 697         rc = ioctl(vcpu->fd, KVM_SET_CPUID2, cpuid);
 698         TEST_ASSERT(rc == 0, "KVM_SET_CPUID2 failed, rc: %i errno: %i",
 699                     rc, errno);
 700
 701 }
 702 /* Create a VM with reasonable defaults
 703  *
 704  * Input Args:
 705  *   vcpuid - The id of the single VCPU to add to the VM.
 706  *   extra_mem_pages - The size of extra memories to add (this will
 707  *                     decide how much extra space we will need to
 708  *                     setup the page tables using mem slot 0)
 709  *   guest_code - The vCPU's entry point
 710  *
 711  * Output Args: None
 712  *
 713  * Return:
 714  *   Pointer to opaque structure that describes the created VM.
 715  */
 716 struct kvm_vm *vm_create_default(uint32_t vcpuid, uint64_t extra_mem_pages,
 717                                  void *guest_code)
 718 {
 719         struct kvm_vm *vm;
 720         /*
 721          * For x86 the maximum page table size for a memory region
 722          * will be when only 4K pages are used.  In that case the
 723          * total extra size for page tables (for extra N pages) will
 724          * be: N/512+N/512^2+N/512^3+... which is definitely smaller
 725          * than N/512*2.
 726          */
 727         uint64_t extra_pg_pages = extra_mem_pages / 512 * 2;
 728
 729         /* Create VM */
 730         vm = vm_create(VM_MODE_FLAT48PG,
 731                        DEFAULT_GUEST_PHY_PAGES + extra_pg_pages,
 732                        O_RDWR);
 733
 734         /* Setup guest code */
 735         kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
 736
 737         /* Setup IRQ Chip */
 738         vm_create_irqchip(vm);
 739
 740         /* Add the first vCPU. */
 741         vm_vcpu_add_default(vm, vcpuid, guest_code);
 742
 743         return vm;
 744 }
 745
 746 struct kvm_x86_state {
 747         struct kvm_vcpu_events events;
 748         struct kvm_mp_state mp_state;
 749         struct kvm_regs regs;
 750         struct kvm_xsave xsave;
 751         struct kvm_xcrs xcrs;
 752         struct kvm_sregs sregs;
 753         struct kvm_debugregs debugregs;
 754         union {
 755                 struct kvm_nested_state nested;
 756                 char nested_[16384];
 757         };
 758         struct kvm_msrs msrs;
 759 };
 760
 761 static int kvm_get_num_msrs(struct kvm_vm *vm)
 762 {
 763         struct kvm_msr_list nmsrs;
 764         int r;
 765
 766         nmsrs.nmsrs = 0;
 767         r = ioctl(vm->kvm_fd, KVM_GET_MSR_INDEX_LIST, &nmsrs);
 768         TEST_ASSERT(r == -1 && errno == E2BIG, "Unexpected result from KVM_GET_MSR_INDEX_LIST probe, r: %i",
 769                 r);
 770
 771         return nmsrs.nmsrs;
 772 }
 773
 774 struct kvm_x86_state *vcpu_save_state(struct kvm_vm *vm, uint32_t vcpuid)
 775 {
 776         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
 777         struct kvm_msr_list *list;
 778         struct kvm_x86_state *state;
 779         int nmsrs, r, i;
 780         static int nested_size = -1;
 781
 782         if (nested_size == -1) {
 783                 nested_size = kvm_check_cap(KVM_CAP_NESTED_STATE);
 784                 TEST_ASSERT(nested_size <= sizeof(state->nested_),
 785                             "Nested state size too big, %i > %zi",
 786                             nested_size, sizeof(state->nested_));
 787         }
 788
 789         nmsrs = kvm_get_num_msrs(vm);
 790         list = malloc(sizeof(*list) + nmsrs * sizeof(list->indices[0]));
 791         list->nmsrs = nmsrs;
 792         r = ioctl(vm->kvm_fd, KVM_GET_MSR_INDEX_LIST, list);
 793         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MSR_INDEX_LIST, r: %i",
 794                 r);
 795
 796         state = malloc(sizeof(*state) + nmsrs * sizeof(state->msrs.entries[0]));
 797         r = ioctl(vcpu->fd, KVM_GET_VCPU_EVENTS, &state->events);
 798         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_VCPU_EVENTS, r: %i",
 799                 r);
 800
 801         r = ioctl(vcpu->fd, KVM_GET_MP_STATE, &state->mp_state);
 802         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_MP_STATE, r: %i",
 803                 r);
 804
 805         r = ioctl(vcpu->fd, KVM_GET_REGS, &state->regs);
 806         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_REGS, r: %i",
 807                 r);
 808
 809         r = ioctl(vcpu->fd, KVM_GET_XSAVE, &state->xsave);
 810         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_XSAVE, r: %i",
 811                 r);
 812
 813         if (kvm_check_cap(KVM_CAP_XCRS)) {
 814                 r = ioctl(vcpu->fd, KVM_GET_XCRS, &state->xcrs);
 815                 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_XCRS, r: %i",
 816                             r);
 817         }
 818
 819         r = ioctl(vcpu->fd, KVM_GET_SREGS, &state->sregs);
 820         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_SREGS, r: %i",
 821                 r);
 822
 823         if (nested_size) {
 824                 state->nested.size = sizeof(state->nested_);
 825                 r = ioctl(vcpu->fd, KVM_GET_NESTED_STATE, &state->nested);
 826                 TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_NESTED_STATE, r: %i",
 827                         r);
 828                 TEST_ASSERT(state->nested.size <= nested_size,
 829                         "Nested state size too big, %i (KVM_CHECK_CAP gave %i)",
 830                         state->nested.size, nested_size);
 831         } else
 832                 state->nested.size = 0;
 833
 834         state->msrs.nmsrs = nmsrs;
 835         for (i = 0; i < nmsrs; i++)
 836                 state->msrs.entries[i].index = list->indices[i];
 837         r = ioctl(vcpu->fd, KVM_GET_MSRS, &state->msrs);
 838         TEST_ASSERT(r == nmsrs, "Unexpected result from KVM_GET_MSRS, r: %i (failed at %x)",
 839                 r, r == nmsrs ? -1 : list->indices[r]);
 840
 841         r = ioctl(vcpu->fd, KVM_GET_DEBUGREGS, &state->debugregs);
 842         TEST_ASSERT(r == 0, "Unexpected result from KVM_GET_DEBUGREGS, r: %i",
 843                 r);
 844
 845         free(list);
 846         return state;
 847 }
 848
 849 void vcpu_load_state(struct kvm_vm *vm, uint32_t vcpuid, struct kvm_x86_state *state)
 850 {
 851         struct vcpu *vcpu = vcpu_find(vm, vcpuid);
 852         int r;
 853
 854         if (state->nested.size) {
 855                 r = ioctl(vcpu->fd, KVM_SET_NESTED_STATE, &state->nested);
 856                 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_NESTED_STATE, r: %i",
 857                         r);
 858         }
 859
 860         r = ioctl(vcpu->fd, KVM_SET_XSAVE, &state->xsave);
 861         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_XSAVE, r: %i",
 862                 r);
 863
 864         if (kvm_check_cap(KVM_CAP_XCRS)) {
 865                 r = ioctl(vcpu->fd, KVM_SET_XCRS, &state->xcrs);
 866                 TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_XCRS, r: %i",
 867                             r);
 868         }
 869
 870         r = ioctl(vcpu->fd, KVM_SET_SREGS, &state->sregs);
 871         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_SREGS, r: %i",
 872                 r);
 873
 874         r = ioctl(vcpu->fd, KVM_SET_MSRS, &state->msrs);
 875         TEST_ASSERT(r == state->msrs.nmsrs, "Unexpected result from KVM_SET_MSRS, r: %i (failed at %x)",
 876                 r, r == state->msrs.nmsrs ? -1 : state->msrs.entries[r].index);
 877
 878         r = ioctl(vcpu->fd, KVM_SET_VCPU_EVENTS, &state->events);
 879         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_VCPU_EVENTS, r: %i",
 880                 r);
 881
 882         r = ioctl(vcpu->fd, KVM_SET_MP_STATE, &state->mp_state);
 883         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_MP_STATE, r: %i",
 884                 r);
 885
 886         r = ioctl(vcpu->fd, KVM_SET_DEBUGREGS, &state->debugregs);
 887         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_DEBUGREGS, r: %i",
 888                 r);
 889
 890         r = ioctl(vcpu->fd, KVM_SET_REGS, &state->regs);
 891         TEST_ASSERT(r == 0, "Unexpected result from KVM_SET_REGS, r: %i",
 892                 r);
 893 }